Hacker News new | past | comments | ask | show | jobs | submit login
Ask HN: What startups are working on hard, technically challenging problems?
155 points by mtae on March 21, 2015 | hide | past | favorite | 170 comments
In the article, "Silicon Valley's Youth Problem" [1] the author mentions Meraki (now Cisco-Meraki) as an example of a startup working on advances in technology rather than the latest web app.

Do you know of other companies that fit the description?

Follow up: Are they (you?) hiring?

[1] http://www.nytimes.com/2014/03/16/magazine/silicon-valleys-youth-problem.html

There are several startups attempting practical nuclear fusion, including:

Helion - has investment from ycombinator, attempting D-D/D-He3 fusion, which would produce only 6% of its energy as neutron radiation.

General Fusion - D-T fusion, but with a clever design that pretty much solves material issues caused by hard neutron radiation. Fusion happens in a spinning vat of molten lead, compressing plasmas with acoustic shock waves from steam-driven pistons. Jeff Bezos is an investor.

Tri-Alpha - the biggest of the fusion startups, quite secretive, with about 30 Ph.Ds, 150 employees, and over $150 million invested. Investors include Goldman Sachs and Paul Allen. Attempting boron fusion, which would produce less than 1% of its energy as neutron radiation.

LPP - the smallest, only about $4 million invested, but might not need more to complete its experiments. Also attempting boron fusion, from a reactor that fits in a small room. About to start a new round of experiments using a reactor core carved from solid tungsten, which they think will boost output dramatically by removing plasma impurities.

Those are the ones that I know have funding. EMC2, the polywell company, is looking for investors now that Navy funding has ended. Non-startups working on alternative fusion include:

Lockheed - this has gotten a lot of press

Sandia - repurposing the Z-machine to attempt net-gain fusion, after simulations showed they could hit breakeven with their existing machine, and 100x to 1000x gain with a 2-3x increase in input power. Very cheap since the Z-machine already existed, and things were going well last I heard.

UW's dynomak project - the most conventional of all these, similar to tokamak but does away with big external superconductors, which makes the reactor ten times smaller and cheaper. Needs $10 million to test whether the idea will scale.

I place higher value on the executive summaries given by the parent than root level links. And you're sending a bill? You must be a management consultant.

A super incomplete list:


- PipelineDB

- Snowflake (Computing)

Internet of Things / Communications:

- Helium


- Pneubotics

- Kuka, namely the research department

Autonomous Systems / "Self Driving Car" et al.:

- Kiva

- Anki

Computer-Vision / VR based:

- Jaunt

- Oculus VR (especially the 'research' department)


- Blueriver http://www.bluerivert.com

(there are many more super-interesting companies in this area!)


- Mill Computing; though quite dubious

- D-Wave

and than there is Microsoft Research working on super interesting stuff in programming languages, computer architecture (FPGAs).

Additionally, I believe really challenging problems will alwyas be coming from creative people, companies in that area; such as Pixar, architecture, and design (keywords, just to give a start: generative {design, art, ...}).

Hope this helps!

I'd say D-wave is at least as dubious as Mill Computing.

I agree but only due to my lack of knowledge in both areas -- and I only have second knowledge by attending a talk of a former D-Wave user and "programmer". That is, and contrasting Mill Computing, D-Wave has implementations in the field.

I'd put in Rigetti Computing in the same category as D-Wave (I'll let you decide how dubious it is).

Oculus isn't a start-up anymore. It's a division of a $100 billion company.

Is Kuka really a 'startup'? They are very well established among the industrial robotics scene.

That's true.

What a name!..

Yes, they're obviously not Scandinavian (where the name means male genitalia, plural).

Funfact: in the northern parts of Norway, calling a policeman a "horses penis" is not illegal, as that is a somewhat common thing to call another person. Yes, this was tested in court in 2008.

I have a theory that if a word is less than 5 letters long, it probably means male genitalia in a foreign language.

Interestingly enough, the number of <5 letter words in English is of the order of 10-15 000 [1], while the number of languages is 7000. Since most languages (I guess) have more than one word for male genitals, it is technically possible for all English <5 letter words to mean precisely that.

[1] http://norvig.com/mayzner.html

Funny coincidence, in some Spanish speaking countries kuka refers to the female genitalia. I have never heard kuka used in Swedish though. It does mean who in Finnish.

Any hints on where I would be able to find more in the Agriculture category?

A few off the top of my head:

  - Trimble
  - Granular
  - Farm Logs
  - The Climate Corporation
I work at Climate, and we work on some pretty cool science based projects to help farmers make better decisions.

We are hiring for a variety of positions you can check out here: http://hire.jobvite.com/CompanyJobs/Careers.aspx?nl=1&k=JobL...

If you (or anyone) is interested in a role on the remote sensing team (satellite/drone suff) or the Climatology team, you can reach out to me directly at skhalsa@climate.com

Climate is owned by Monsanto (market cap USD 56 billion).

Terminal.com is writing next generation virtualization. This includes live-migration without hypervisors (already in production) and live-resizing (also in production).

I think we'd like to think we're working on the ugly bits of infrastructure people don't care about.

This includes a distributed file system optimized for speed and storing machine state like Github, software defined networking for IP migration across metal, and other abstractions for making devops easier.

We're looking for people who want to think about hard computer science problems like managing stateful systems more intelligently (think auto-failure detection and recovery and less I/O intensive WAN replication strategies).

I just heard of Terminal.com right here, and trying it out for a spin, but it looks like the disk performance is really low, is this expected for lower tier or trial instances?

This is really interesting. I have never heard of Terminal.com before and I'm now deeply interested. Do you hire remote or sponsor visa?

Feel free to ping me on the email address in my profile. Be forewarned, we have a pretty serious technical interview for all candidates. I would say it's not easy to get a technical job at Terminal.

Thank you. I pinged you ;)

I submitted this to slashdot a few weeks ago, but it got declined:

"For some time now, I have been following the "startup scene" and frankly, I am left with a sense of dismay. How many of the startups actually do anything of any real value to mankind? It seems to me that the startup ideas just keep getting more ridiculous and stupid by the day and I think I would go as far as to call the whole thing deeply broken.

I am not going to name any specific startup, but I would like to ask the readers of Slashdot a question.

I know this is not how the world works, but I am still curious to know what kind of ideas would prosper if the primary aim of a startup was not to make as much money as possible as quickly as possible. So, if your startup idea would be judged by the amount of good it would to mankind, what would it be?"

Where has this concept of "world changing is important, and a business must be of real value to mankind" come from in recent years? Who gives a shit how incredibly, ultimately, world changingly seriously important the work is of a business?

Since when did this become some sort of measure of the worth of a tech company?

Sounds like misguided hippy shit to me.

Bill Gates, Larry Ellison and Steve Jobs didn't set out to change the world and "do something incredibly important for mankind", despite what the message might have become in later years. They all set out to do things they were interested in, pursuing technology for its own sake, as most people were back in the 1970's.

These days startups seem to need to feel that to justify their existence they need to be "changing the world" or "doing something important for mankind" or "giving back to the community". There's no need to justify the existence of a business. Do it for the cash, do it for fun, do it to scratch your sense of ambition, do it because you think it's a good idea, even if some pompous git stands on a high horse and looks to the sky and proclaims "this business is not changing mankind for the better, I am dismayed at the trivial nature of this endeavour, I deem it of little value".

It's exactly because all these startups pretend to be "changing the world" that they become measured by these promises. And obviously most come up short.

But that doesn't need to be a bad thing. Paying lip service to "real" progress, especially when it concerns poverty/environment/democracy etc, may actually lead to actual progress down the road.

Where has this need come from for startups to feel illegitimate unless they are changing the world, giving back to the community, bringing world peace blah blah

Well, if you start pitching this to everybody and their mom too, people will start to measure you up based on this.

At first it was pretty rational to present yourself in a way that gets you that 15 minutes. Now, I don't know, but HN is full of startups trying to revolutionize something instead of incrementally "disrupting" a set of incumbents.

Interesting. I don't think there is such an explicit need. I blame Google and their "Don't be Evil" tagline.

On the other hand, two other popular startup mantras "making products that people love" or "fixing a problem that people have" are quite easy to spin to be worthy and good.

Well, that's ok, we all see things differently. And I guess, I don't have a problem with that. If that is what you want to do, go for it. But I would rather not be called some pompous git, I really think that is as far from what I am as possibly can be. Ok, I can be an asshole sometimes, but I try not to be.

Trying to find uses for the internet? Oh well. If you think it is pompous to "judge the startups you see against some measure of "importance to mankind"" I think there is a problem, and you are probably part of it. I don't want to argue. Just go make lots of money and find some use for the internet and all is well.

It's pompous to judge the startups you see against some measure of "importance to mankind". Sadly many startups seem to judge themselves against the same standard and seem to have some sense of guilt if they aren't creating clean water for millions in the desert.

Perhaps you should rejoice in the fucking incredible explosion of energy and creativity and effort that the entire world is putting into trying to find uses for the Internet.

Just relax with your hippy ideals. Have fun.

Take heart, and don't lose too much cynicism. People feel like they have to include all this "hippy shit" because of the exact reason this question was asked - "if only I could find a good company who's mission I believe in!"

Too many companies making incremental BS apps or IT pipeline tech that delivers cat pictures 0.1 ms faster. And both groups ("we're bringing Utopia to Earth", "we'd mug you for cash") are still doin it cause it looks like it makes money, or people advise them they need to say these things to make money, or it'll get chicks / dudes, whatever - no matter what they say out loud. "Silicon Valley" is pretty spot on in that regard.

It's just a pendelum swing. In a little while we'll be back to "greed is good" while we help people 'connect' with high school friends they never actually liked.

> Who gives a shit how incredibly, ultimately, world changingly seriously important the work is of a business?

I don't give a shit about how world-changing a business is, but I do care that companies should be driven for more than just making money. Most companies aim to be profitable, but if that's the only goal then they're primed for corruption (even on a minor scale). Who cares so long as profits keep growing, dog eat dog, right?

There's a difference between 'Can I do something?' and 'Should I do something?', I see no harm in wanting companies that consider the latter.

Wasn't putting "a computer on every desk and in every home" Microsoft's mission?

I think a lot of startups are solving important problems that might not be obvious.

Any Musk startup - Tesla (electric vs. oil consumption), SpaceX (building space "colonies"), SolarCity (alternate energy), Paypal (payments).

You also have Uber (limiting cars on the street & oil consumption), AirBnb (better utilization of current buildings rather than building additional hotels). Even stuff like meerkat, etc could potentially be used in lieu of expensive meetings across country (further travel & ancillary costs).

I do agree that "change the world" is an overprevalent mission statement & too broad to be useful, but many are solving problems in a way that might not be obvious at first.

AirBnB is more about exploiting cost differentials by avoiding regulation, isn't it? When i use it, and the people I know making money on it, are exactly taking advantage of this versus "better utilization". In fact, it's actually the opposite, at least in places like SF, because units that would otherwise house residents are now being used to house hotel guests.

> AirBnB is more about exploiting cost differentials by avoiding regulation, isn't it?

That's a story one-side uses and in SF that can certainly be the case.

But, in some cases, I think people would rather stay in places that are less "developed" or taken over by hotels and stay at local BnBs.

For the purpose of "better utilization" there is no difference between residents and tourists. Airbnb puts more people into the same space.

If Meerkat is a productivity enhancer, then surely Yo is one as well - when people don't have to write text messages, it saves time for something else.

Live streaming tech can certainly use push forwards.

Communication methods that people laugh at (AIM instant messaging) can be reappropriated into something useful (hipchat, slack) years later.

Oh, FFS. Tesla has done a fine job at showing electric cars are feasible, for sure. But without real innovation and progress in emission-free baseload electricity generation (i.e. fission or fusion) it has negligible effect on reducing GHG emissions.

Anyone care to explain why I get downvoted for stating an accepted fact? Take e.g. the IEA Blue Map scenario (50% reduction in GHG emissions by 2050), which depends heavily on electric vehicles. It also requires a 320% increase in nuclear power generation.

I feel like at least in this thread, you're forcing a square peg in a round hole. It's asking for technically challenging startups- there are plenty of startups doing amazing, world-changing work that isn't technically challenging, but more socially challenging.

You are probably correct. I just get a little frustrated sometimes. Sorry :)

No worries. It's easy to get disillusioned when the tools to surround yourself with better options aren't very well developed.

This is because startups are not supposed to be working on these "seemingly big" problems. If a problem looks big and looks like it potentially has a huge market, every other large company in the field would be already working on this. As a startup, it is stupid to try to tackle something that obviously big. That's why most successful startups start out as something that doesn't look threatening. Of course, there are genuinely ridiculous ones like "uber for clowns", "kickstarter for baristas", "airbnb for prostitutes", etc. but I can guarantee if you look back in a decade, what actually changed the world will have been what you now consider ridiculous.

In Berlin, there is the 'Social Impact Hub'


of which I am not affiliated with nor won't endorse here in any way. Sadly, in Germany at least, startups that aim for maximum positive impact and less so profit, are called 'social startups' which outside of Germany probably has a completely different connotation.

The news websites I usually frequent, are not very full of news; they are rather SV & VC company theory megaphones. Avoiding those websites sometimes leads one to people with a bit different entrepreneurial mindset.

This having said, would you work for such a company? And which compromises would you accept (on the ethics-, but also wage-scale)?

If you are asking me, my answer is yes. I do not care very much for money. If I make enough so that I can eat and not have to live in complete misery, I am ok.

Ethics, that is harder. I guess if you look hard enough you can find problems with anything, but that's probably just the way the world is. If it does more good than harm, it's ok in my book.

Value is ultimately a subjective term. Your perception of value is different than my perception of value, which is different than millions if not billions of other people's perception of value.

There are countless startups which were perceived as having little value only to be praised and hugely successful years later, yet nothing changed– only people's perceptions.

I guess my point is that before asking a question like that, we would have to define what you constitute as "doing anything of real value to mankind". And who's to say your definition of value is the right one anyway?

A few ideas below. You mention startups trying "to make as much money as possible as quickly as possible" - I personally quite strongly believe that companies that focus on long-term profitability also happen to do a lot of good for mankind.

- cure for cancer

- cure for malaria

- cure for aging

- cheap fossil fuel extraction from low-quality deposits

- safer nuclear fission

- nuclear fusion

- lower-cost space travel

- space elevator

- space colonisation

- intelligence augmentation

- safe artificial intelligence

- desktop nanofactories

- nano-machines

Of these, fission/fusion/fossil fuels are the only ones start ups might touch. And even there, it depends on a lot of physics turning out favorably.

At least the Helion guys believe they have a good shot at it http://www.helionenergy.com/

See my post above about what we're trying yo achieve with Graphistry. Unlocking 20 years of infoviz research by letting anyone run on 1000 real-time GPUs helps move some of the most important sectors forward, eg, biology, security, and finance.

What do you mean, real-time GPUs? And: what interconnect are you running on? How does your scaling look? Is this just for embarrasingly parallel stuff? Just curious; I'm running multi-GPUs myself for molecular dynamics.

Imagine moving a slider in your viz to change a filter, physics setting, etc. and have cluster start to immediately feed back new results.

We started on building fast-start multitenant access to single GPUs and approaching peak on those (full-GPU barnes hut, 10X over Keshav's work). We're now focusing on distributing, and as we are more interested on running on many GPUs for scale out, focusing on communication avoiding. This makes a path to giving companies time on 1000 GPUs (think Pixar-levels of compute) rather than shipping small 8 GPU boxes with infiniband. Via elasticity and time sharing, the analyst hour pricing is unprecedented.

The titan guys run on 20,000 GPUs for similar astronomy codes, so doable. We're making it in more accessible, big-team, and analyst-focused ways. E.g., load, interactively analyze with smart defaults & streamlined common paths, export/report, and share.

Ok, I see, so your approach makes sense for problems where throwing some GPUs at it gives you a solution in O(10) seconds? Sounds nice if you know you hav problems that fit into that category.

I found the Graphistry webpage lacking in answering the question "which specific problem does this solve?" Infoviz is too broad/vague.

Yep, except think magnitudes bigger & faster. Right now, we're applying this to visual graph analytics problems in a few key industries. If you do infoviz, email us and I'm happy to share more!

Sandstorm.io (https://sandstorm.io) is working on turning server infrastructure and security upside-down without the need to rewrite applications. The goal is for distributed infrastructure and security to be as easy as possible for end users to manage, and as hard as possible for applications to screw up. The model we're building is capability-based; previous capability-based research systems have usually required rewriting apps to fit the model, but ours does not. This leads to a lot of interesting computer science, systems engineering, security, and UX problems, both in designing a new model that is practical and just-in-time reverse-engineering of application intent in order to seamlessly fit it into the new model.

Possibly-interesting technical posts:








Sandstorm is not currently actively hiring. But it is an open source project if you're interested in contributing. See:


Wondering why you're not actively hiring?

We only just raised a seed round[0], which only gives us money to hire a couple of people, and we already have more really awesome candidates than we have budget to hire. So basically we're all set on hiring until Series A (2016?). You can feel free to e-mail jobs@sandstorm.io if you think you're an exceptional case, but probably the answer will be: "Sorry, we don't have budget or bandwidth to interview more people right now. :("

[0] https://blog.sandstorm.io/news/2015-01-15-sandstorm-1.3M-see...

MapD (http://www.mapd.com) is building a hyper-interactive big data analytics and visualization platform running on multiple GPUs/CPUs. We basically do every performance trick in the book (like compiling our queries via LLVM and caching hot data on the GPU) to do this. The end result is the ability to scan billions of rows of data in milliseconds at rates greater than 2 terabytes/second per server when running on 8 Nvidia K80s (although we also get great results on laptops and even ARM cores.) And not only can you query the data with SQL but you can visualize it or feed it into machine learning algorithms without copies because it is already distributed across all available compute devices.

Where did your cofounder go? Why is half your company non-engineering (per your team photos)?

Tech looks good, company looks shaky.

Actually all of our team codes, we just wear other hats at times. Our Director of Business Development has a PhD in GPU algorithm acceleration and I (the ceo) coded the original version of MapD and still play full-time engineer as much as possible. I can see why one might get the impression you got from the website though - perhaps we should add a bit more background on ourselves.

At Graphistry, we are powering nextgen visual analysis through real-time GPU clusters that stream right into your browser. For example, we cracked a botnet a couple days ago using it. Visually exploring millions of data points in real-time is a hard problem for both infoviz design (hairballs!) and high performance programming (1000 node GPU clusters), but unblocks some of the biggest industries (finding criminals, avoiding economic meltdown, datacenter rootcause analysis, ...).

And yes, we're looking for the right data viz folks to help us harness all this performance and help data analysts make sense of it all.

> we cracked a botnet a couple days ago using it

Do elaborate...

We missed at least 60,000 bots in our first analysis so will write up after we get the rest of the relevant input data & rerun. If you do infoviz, please contact and I can explain more :)

I love how due to the nature of the Internet that huhtenberg may be the owner of that botnet. ;-)

(note: not saying he is, obviously. just making a point we should all be aware or when communicating with strangers online.)

If he takes it down without authorities forcing him to, great!

Kryptnostic seems to be working on bringing usable fully homomorphic encryption to market. If they actually have something real there, that could be game-changing for cloud privacy:


It's a shame Google isn't focused more on this, too. They should be having a Google X lab just for that, and invest at least as much as they currently invest in quantum computing. They have a lot to gain from it. Infringing on people's privacy is currently Google's biggest PR issue, and it's not going to go away. It will only become a bigger issue in the future. Without this, Google could perhaps return to being almost as liked as it used to be (there's still the issue of bullying others in the search engine, of course).

The first company that can crack the full operational cycle of homomorphic crypto will get a license to print money. As long as they don't squander the first-mover advantage, it will take quite a long time for the eventual second-mover to take over.

It's not just doing the computations in an untrusted cloud environment, although that's a big promise. I would say that an even bigger part will be the eventual ability to run computations on multi-party data sets. If you can provide a provably secure zero-knowledge service that still manages to compute results from multiple sources of opaque data, you will have every single financial institution and insurance giant banging on your door.

I work in the data analytics space, so I'll focus there since many of the other answers give a good, broad set of companies to consider.

* Trifacta (http://www.trifacta.com/) are dealing with the very grungy problem of data transformation. Their approach is heavily UX-centric, with built-in predictive capabilities that learn what you are doing as you try to transform disparate data sets to meet your needs.

* Segment (http://www.segment.com) acts as a data router, allowing you to implement significantly less data plumbing in your application while allowing you to deliver your data to many different analytics tools.

* Jut (http://www.jut.io) is a full-stack approach to building a hub for streaming data. They take any operations data (logs, metrics, alerts, events) as inputs, manage storage and analysis, and have creating a framework for streaming visualizations as well. Technologies include an in-browser, retargetable compiler, a streaming analytics layer, storage (elastic search and cassandra), A d3-based visualization framework designed for 3rd party add-ons, and a simpler way to manage large-scale data called hybrid SaaS. (disclosure: I work here.)

* Databricks (http://www.databricks.com) has implemented apache Spark as a service.

At Packetzoom we're building up a networking protocol specifically designed for mobile networks. We've discarded the TCP/HTTP stack completely and started from scratch. The idea was to go back to the drawing board (to the lowest level possible) and ask this question:

How would network protocols on mobile work if they were being designed from scratch today?

And that's what we've done. Results so far are pretty amazing (yes, we're out in production with private beta customers).

And oh... sorry but we're not hiring right now. We have a small, tight-knit team and we plan to hire very slowly with only the exact right fit of people.

I have an honest question: what do you answer when the obvious comparison to (and lack of industry adoption of) SCTP is put forward?

In-protocol multiplexing must be pretty high up the list of necessary features, but how about transparent multi-network roaming? Not just over mobile operator networks, I'm talking about hopping between different transports.

Let me answer your latter questions first:

In-protocol multiplexing must be pretty high up the list of necessary features

Yes :-)

but how about transparent multi-network roaming

Yes :-)

I'm talking about hopping between different transports.

Yes :-)

(I was thinking of writing a much more detailed answer to above, but it's so detailed that it would be far better as a blog post... I'll post the link here for a more inclusive discussion soon)

Now going back to your first question:

"what do you answer when the obvious comparison to (and lack of industry adoption of) SCTP is put forward?"

Oh... where to start.. OK. First off the comparison is not all that obvious. SCTP was never built for the mobile use case (4 way handshake??). There are some improvements over TCP (reliable, out-of-order message based delivery with multihoming are all good things), but there's a fatal flaw that prevents any serious use of this protocol. It's not TCP or UDP!! And there are umpteen middle-boxes on the internet who won't let it through. That's all you need to know to explain the lack of industry adoption of SCTP.

So we built our protocol on top of UDP (which we basically treat as a proxy for raw IP). Luckily for the mobile users, UDP based custom protocols are now well baked into various standard applications (DNS, VOIP etc.). A vast majority of middleboxes support it and the trend is in the right direction -- every new middle box has to support UDP passthrough/NAT by default barring exceptional circumstances.

Now, how do we expect our new protocol to get adoption. The answer is, we don't. We're not just building a protocol. We're building a turnkey service for mobile app developers. Our business case doesn't involve convincing anyone of the technical superiority of the protocol. We simply sell a superior user experience with minimal developer effort. And you'll learn as we get more open in the coming few weeks (wink, wink), it's not at all a hard-sell :-)

Feel free to connect with me directly or just continue the discussion here if you have more questions. I'm always happy to engage skeptical observers ;-)

Let's start with the encouragement: please, please put the technical details into a blog post and publish it on HN. This is the kind of stuff I've always[tm] been interested about.

> umpteen middle-boxes on the internet who won't let it through. That's all you need to know to explain the lack of industry adoption of SCTP.

Heh, I've dealt with IPSec. You don't have to convince me on the futility of getting new protocols ("the magic number on IP header") approved for end to end delivery. :) For better or worse, UDP encapsulation is the sensible thing.

I used SCTP as a fairly well known example of a protocol that has many of the modern technical requirements built in. As far as protocol features go, it is a pretty good yardstick.

Now, as for "turnkey service"... I would guess you're positing yourself as the platform provider. You need to control the server endpoints to work with multi-homing; you'll have to provide and maintain the mobile app libraries, along with best practices documentation and aids for debugging; you probably want to provide a patch for wireshark (dissector); and then there's the security to think about. That's just off the top of my head.

I am genuinely interested how you have solved the problems and where you have looked for inspiration. Because this is aimed at mobile developers, unreliable latency will be a major factor. That's something you do not have control over - maybe the protocol library can help a bit, and multi-homing already mandates graceful handling of concurrent retransmits. But mobile networks are not just unreliable, they are all too often outright crappy.

Can one ever work around that?

Now, as for "turnkey service"... I would guess you're positing yourself as the platform provider. You need to control the server endpoints to work with multi-homing; you'll have to provide and maintain the mobile app libraries, along with best practices documentation and aids for debugging; you probably want to provide a patch for wireshark (dissector); and then there's the security to think about. That's just off the top of my head.

That's a good list of issues. And yes, we've thought about each of those issues and have had enough success that a few pretty rigorous customers are currently using us in live apps.

Because this is aimed at mobile developers, unreliable latency will be a major factor. That's something you do not have control over - maybe the protocol library can help a bit, and multi-homing already mandates graceful handling of concurrent retransmits. But mobile networks are not just unreliable, they are all too often outright crappy.

Well the general issue of lossy connections, unpredictable latency and outright disconnections and reconnections (often with a different IP address) is the main impetus behind designing a brand new protocol. TCP does a woeful job here and any new approaches that limit themselves to just tweaking server side TCP are DOA when it comes to solving those issues. We'll be going more public with some of this info soon (in a couple of weeks).

Btw, checked out Cricket WC odds on Smarkets. I think you guys are spot-on on the second Semi Finals ;-)

> TCP does a woeful job

Not just TCP. Network operators are guilty too. There are the network side megalith buffers that are trying hard to "optimise" TCP transfers and just murder latency. Then, precisely because TCP is a stateful protocol and clients have the retransmission logic built in, the operators also have the option of applying something like RED at the earliest sign of network congestion. False positive or not.

(Note: I have a friend who works on the other side of the table and runs a team of engineers responsible for installing new mobile network equipment in the field. The use of mobile networks grows faster than the operators and their contractors can add capacity. Which means the situation is unlikely to improve soon.)

As you mentioned earlier, UDP at least has the distinction of being used in real-time and streaming protocols => operators and network gear are more reluctanct to mess with it. If you think you can get away with it, you could also assign multiple ports to any single session and apply at-least-once logic to the transmission - or to put it in layman's terms: blast and spam the network, in order to get even a single datagram out without excess delay. Do that often enough and you can expect the operators to crack down on it, though...

As for work: thanks :-)

If you're not hiring, just don't mention open positions. My entire impression of your company is now "the CEO is a dick."

Sorry.. where were the open positions mentioned?

Oscar Health in NYC is fixing the health insurance industry by building a tech and data driven insurance company from the ground up. It's an extremely complex, highly regulated industry that is in need of fixing. There are many startups that try to address pieces of the broken health care system, but Oscar is the only startup trying to solve it by doing it all. Engineering talent from Facebook, Google, Tumblr, Spotify, Apple, and more.: https://www.hioscar.com

The USA shouldn't have such a terrible health care system. Let's fix it. They are hiring all sorts of talent: https://www.hioscar.com/jobs

Fully cloud based, heavy aurora/mesos shop, kafka, hbase, redshift, mysql, python, flask, and an abundance of data analytics.

No doubt health insurance is hard problem, but it's not technically hard. OP asked for examples of company's pushing the limits of advanced technology, and Oscar is not one.

It's not? Why do you feel that way?

It's not solving difficult problems in algorithms, systems engineering, hardware, ML, AI, etc.

Nothing wrong with that. But using today's technology in web and mobile to build a highly-usable customer experience, is not what OP meant by hard technical problems.

I'd like to correct you on that, Oscar is not just a pretty design smacked on-top of an insurance company. It's the entire insurance stack redesigned from the ground up.

Algorithms and ML are all heavily used in analyzing member data to improve the lives of people.

We're engineering systems that give us real-time feedback on the insurance system as a whole, rather than the usual "30-60 days". Building a claims system from scratch is not an easy feat.

If you'd like to learn more about what we're doing i'd be happy to demo it for you.

Don't be fooled by our pretty and simplified website. It's all of the hard engineering efforts that you don't see that make the simplified experience possible. :)

Good luck! I hope you guys do well and deliver an awesome product :-)

According to basically every career page of every startup, every startup ;)

You're right, I should have been more specific in the title. Although I do think the description gives a better idea of the kind of startup I was looking for.

I'll play devils advocate since no one else is. Why does the company's product(s) have to be the advanced technology? The successful "latest web app" startups have created tons of tech they build on top of. Scale leads to technology.

At Airware, we are doing hardware, embedded systems, flight controls, desktop development, and web services. We deal with every type of data that could be captured by a 5-50lb drone aircraft and processed or analyzed: images, video, logs, streaming telemetry, etc. Problem spaces include autonomy, machine vision, and distributed computing (both on the vehicle and in the cloud).

Yes, we are hiring: http://www.airware.com/careers

Are you hiring interns (for the fall)?

Skymind is building an open-source, distributed deep-learning framework for Java and Scala.

http://www.skymind.io/ http://deeplearning4j.org/

Plus their scientific computing library http://nd4j.org and machine-learning vectorization library Canova.

At Spire (spire.com) we are building small satellites to monitor the weather and global trade. We're looking to improve the efficiency and safety of shipping as well as help prevent illegal fishing. And we're hiring.

https://UtilityAPI.com/ is making the Twilio for energy.

Humans needs to switch 87% of our energy sources (~450 Quadrillion BTU/yr) from fossil fuels to other source in our lifetimes[1]. That's an unbelievably huge transition, and one of the current bottlenecks is energy data communication. Solar and energy efficiency companies can't communicate efficiently with utilities and vice versa because each utility has a unique, antiquated, manual system for handling energy data. We are trying to fix that by making a universal API that wraps utilities so energy innovation can happen efficiently on both sides of the meter.

Yes, we are hiring: https://angel.co/utilityapi/jobs

[1]: http://www.pvsolarreport.com/the-next-internet/

At Parallel Universe we are working on a full server-side stack (distributed in-memory storage, through the database, and all the way to the communication layer) that is mechanically-sympathetic with modern hardware architecture (most current software architectures actually fight the hardware).

Most of our projects are free software, and yes, we are hiring.

Tachyus - Predictive analytics for the oil & gas industry. They offer solutions for production optimization in cyclic steaming, steamflooding, CO2-flooding, waterflooding, pump optimization, workover prioritization, and shale fracking.


Three startups I want to promote:

1) Onu: A cloud computing framework based on GPUs. I have worked with both of their co-founders and have really high hope for this company in the future of cloud computing market. They are hiring top GPU programmers: http://anticipate.onu.io/

2)Tachyon Nexus: memory-centric distributed storage system. Just announced a couple of days ago by my friend Haoyuan Li from Berkeley AMP lab. http://www.tachyonnexus.com/

3) Mental Canvas: A spin-off from Yale professor Julie Dorsey, they are working on 3D reconstruction from 2D drawing. Very cool stuff and one of my friend is the early employee there: http://www.mentalcanvas.com/

Blockstream is writing the code that will one day run not just the entire financial system, but eventually all forms of digital property, contracts, and services drives thereof. And we're hiring:


^services derived thereof. Stupid phone auto-correct...

I work at a network security startup [1], I can't say much until we come out of stealth mode on Tuesday, except that it's a very challenging engineering problem.

[1] https://www.protectwise.com/

We were working on autonomous stuff, but had to move onto simpler things because people weren't buying.

http://robots-everywhere.com/portfolio/navcom_ai/ (2007)

This reflects your current status I presume? http://robots-everywhere.com/re_site/

Yeah. Was trying to not link that coz I don't want to spamvertise.

I see. Nitpicks: - The twitter icon on the top right points to @spiritplumber which has the name of a "Linda Camezon", a standard issue avatar and only 1 tweet which doesn't signify anything. - The text at the bottom of the website (in green) is unreadable given the background. - I couldn't find a careers section or the location of the company. - The videos are not professionally produced which gives the feel of an amateur effort not an established company.


Periscope (https://www.periscope.io/) is a data visualization tool that automatically and transparently syncs customer data into a huge multitenant data cache that runs queries ~ 150X faster than customers' own databases.

In the long run, it is a data tool that will eliminate the need for data warehouses.

It turns out the hard problem is not running the queries fast, but keeping the cache accurate and up-to-date. Our cache coherence service is probably the piece of code we're most proud of.

And, yes, we're hiring in SF. :) Reach out to me, harry@periscope.io

For a second, I thought this was the streaming video platform that Twitter just bought.

I read this a few months ago. I agree with its interpretation over the current state of the entrepreneur culture amongst the youger cohort of emerging adults. I believe that I am working on a unique problem. One that involves being able to create a behavioral profile on an individual based off what their online behavior (i.e. their public timeline information) and connecting it to how the tend to spend. (categories or unusual trends) Not currently hiring, but open to talk to people interested in it.

MetricWire (https://metricwire.com/).

We are building a platform that enhances the workflow of clinical trials thus making it easier, faster and cheaper on drug companies and safer for the participants involved. We help the researchers using the data collected from the trial to adapt the future stages of the trial thus helping avoid serious adverse effects or making them known to researchers in real time if they are occurring at present.

can you expand on the technical r&d involved? Sounds like a pretty normal user interface & data sharing problem.

In a way it is. So basically our platform allows users to create studies and structure the entire workflow of a trial(in a lot of these they are multi-staged so there are multiple steps that need to be designed). Now after this done, they can follow this workflow on our platform when they are conducting their trial in house. Now for parts where a doctor/researcher is not present(like drug/medication/treatment application), we have a mobile and web app that lets users answer questions/file serious adverse events if they face any, etc.

Now here is the cool part, the above is just the surface. Our team is really passionate about data science, machine learning and artificial intelligence and its applications in real world. So we do something called adaptive clinical trials where we run analytical models on the data that comes in and can forecast when the trajectory of a trial needs to be altered. This means you can catch any serious reactions to drugs, faster.

All in all, to answer the main question, our solution helps keep the costs of clinical trials low and reduces error rate. Being able to run these trials fast and keeping the costs low, allows not only for drugs to get out to the market faster, but also helps reduce the costs of drugs(due to the reduction of costs to process).

I also only explained the application of our tech in clinical trials, but our software is so flexible that it can be applied to any industry such as Consumer Insights, Depression analysis studies, etc.

Not sure if I adequately answered your question.

At Meteor, we're building a complete open-source platform for easily building best-of-class, fast, web and mobile apps. If you've never heard about Meteor, check out the original screencast: https://www.youtube.com/watch?v=fsi0aJ9yr2o (needless to say, lots have changed since then -- most notably a security model :)

Meteor touches all parts of an application stack (client-side templating[1], live database updates[2], a websocket-based sync protocol[3], an "all you need" build tool[4] that can build mobile and web apps, an in-house "transparent reactive programming" library[5], a cloud hosting environment[6] built on Docker and Kubernetes, and the list goes on -- all with a very very strong emphasis on great developer experience).

Learn more about Meteor's subprojects here: http://meteor.com/projects. Most of these projects are current edge new technology -- we develop new systems when the existing open source ones aren't good enough (either because of how they're implemented or because we can supply a better developer experience)

Moreover, almost all of our work is open source which means working here builds your GitHub profile -- always a good thing. Remote is OK. Learn more at http://https://www.meteor.com/jobs/, or happy to answer questions at avital@meteor.com.

[1] https://www.meteor.com/blaze

[2] https://www.meteor.com/livequery

[3] https://www.meteor.com/ddp

[4] https://www.meteor.com/isobuild

[5] https://www.meteor.com/tracker

[6] https://trello.com/c/FMdB7GAu/78-galaxy-managed-production-q...

Have you seen Carbon3D? http://carbon3d.com/

(I'm not affiliated with 'em, just think they're pretty rad)

At FarmLogs (YC W2012) we're using high-precision satellite, tractor, weather, and nutrient data to generate predictions and recommendations for crop growth.

Counsyl is doing great work in genetic testing. https://www.counsyl.com/careers/

From the website for Vicarious

"our mission: build the next generation of A.I. algorithms"


MaidSafe (http://maidsafe.net) recently raised over $8m from one of the first ever cryptocurrency-based rounds of financing. As the company approaches beta launch, they are looking for a front-end Qt Developer to create applications to sit on top of the SAFE Network, a secure and fully decentralized data management service made up by the individual users who contribute storage, computing power and bandwidth to form a world-wide autonomous system.


"Know any #QT (C++) developers who want to help change the world from our Scottish HQ? Please ask them to get in touch: careers@maidsafe.net"


Features of the SAFE Network:

- Self-authentication: users can create accounts and log in from any computer without the need or knowledge of third parties (http://systemdocs.maidsafe.net/content/en/system_components/...).

- Farming: this self-regulating network lets users offer spare drive space and in return delivers anonymous, super-fast internet (http://systemdocs.maidsafe.net/content/en/system_components/...).

- Examples of apps that can be used on the SAFE Network: cloud storage, encrypted messaging, websites, VoIP, social networks and any existing service that runs on the Internet (http://maidsafe.net/applications).

- Safecoin is the currency of the SAFE Network and a mechanism to incentivize and reward farmers and app developers as well as provide access to network services (http://systemdocs.maidsafe.net/content/en/how_it_works/safec...).

If you have any questions about the SAFE Network, please ask them on https://forum.safenetwork.io.

Visidraft is working on making Augmented Reality actually deliver on the promises of presence on mobile devices at the consumer level and give 6 degrees of freedom around virtual objects in the real world context.

We have to make Large scale Simultaneous Localization And Mapping (SLAM) easily computable across existing mobile devices to bring consumers hyper-precise (cm precision) geo-reference. This is actually super hard but also pretty awesome.

Yes we are hiring. Specifically CV experts.




Quobyte. We're building a fault-tolerant scalable distributed storage system for POSIX files, block and object storage. Lots of challenges around distributed systems, architecture and concurrency.

Development is in Berlin, Germany, and we're hiring.


Any plans for considering remote workers?

What's the revenue plan for a business like that?

Software licenses.

Deepmind, working on general AI, and yes, they're hiring.

No longer a startup -- they got bought by Google.


seconded. geo-rectictification is insanely difficult and computationally intensive

I work for Pivotal, which is working on a bunch of problems. We're a startup in that we were founded on investment capital (EMC and GE) and we're growing very fast.

Probably the biggest effort we're making is on Cloud Foundry, an opensource PaaS. Developing scalable production-grade systems for managing and running 12 factor apps is hard. Superficially it looks easy. Once you do anything at scale, it stops being easy, which is where we come in.

We also have teams working on Hadoop, GemFire, antirez is on Redis and Disque, the Spring team works for us and I've kinda lost track of the rest.

I work for Pivotal Labs, the agile development wing founded in 1989 from which the whole company takes its name.

We're hiring across all divisions. Email me: jchester@pivotal.io.

At Second Spectrum we use spatial-temporal pattern recognition to tell stories about sports. We have lots of interesting challenges in machine learning (how to recognize patterns from moving dots), user interface/experience, (how to intelligently display the mountain of information that we gather from a single game or even play), and computer vision (how to augment video to display semantic information from our pattern recognition layer).

And yes, we are hiring in LA, in those areas, as well as for full stack developers. Feel free to reach out at noel@secondspectrum.com or work@secondspectrum.com

At The Climate Corporation we're making science based tools to push forward precision agriculture.

I quite like it, there are lots of hard technical problems. I realised that since I joined ~1 year ago, I haven't worked on any CRUD apps at all yet.

Here are some positions: http://hire.jobvite.com/CompanyJobs/Careers.aspx?nl=1&k=JobL...

Reach out if you have questions: skhalsa@climate.com

At Netra (http://www.netra.io) we are building true visual search. Imagine being able to search video/images using other image(s) or video clips. Yep we are doing that, and eventually Google scale.

We see immense benefits to humanity - from medicine, to environmental science, crime prevention/monitoring....down to the trivial - search millions of your photos for your pet dog, look for all photos where you & Aunt Millie were together etc.

p.s : and yes we are hiring.

At Clarify.io, we're making audio & video searchable. Unlike a general purpose ASR (Automatic Speech Recognition) engine, we perform a bunch of automated analysis to tune the engine and optimize it for the particular language, accent, and a variety of other things. We've found that our accuracy is much higher than a general purpose one.

The founders went through Techstars London 2013 and moved to Austin right after. We raised 1.4M last fall from a mix of angels and VCs.

nvidia is doing amazing work, and many of their employees are in the bay area. 3072 cores for $999 - http://www.geforce.com/.../geforce-gtx.../specifications

The number of companies using this device to solve real problems is growing. Jeremy Howard just started an interesting one - http://www.enlitic.com

How is Nvidia a startup? They were founded in 1993 and have 8,800 employees and revenue over $4B.

Nvidia has done great work with CUDA, but I wonder if AMD/ARM's HSA will make the work of the people using CUDA now much easier in the future - compute with a high-level language that you already know. That's got to be some kind of paradigm change there. I just hope they end up supporting Rust, too.

Compute on the GPU is unlikely to be unified in a meaningful way with the CPU in the near future. The GPU is a massive SIMD machine in contrast to the CPU.

Even if you use the same language for both, the style of code & algorithms are very different.

Check out the AMD APU. We've had good experiences with recent Intel integrated as well due to embedded RAM. You are right though in that we write our code with data parallel and sometimes manycore SIMT in mind.

I worked on the Intel integrated embedded RAM a couple years ago (before release)! It was tough getting the h/w bugs out. Are you using it in Windows or Linux?

Frustratingly, only when we do the client-only demos on our macbooks (iris). Our 'real' version is AWS or dedicated nvidia boxes, where we don't get it. OTOH, this sort of thing makes me excited that we can expect significant HW-driven speedups for our approach for years to come.

Nvidia is a publicly traded company lol.

At JamKazam, we are trying to enable real-time play of music over the internet. While we can't magically make the internet better, we have spent a ton of time getting latency as low as we can in Windows, Mac OSX, Linux, and on custom hardware (still under development).

Just a week ago we released our 'distributed metronome', which lets musicians hear a synchronized metronome regardless of how much internet latency there might be.

We're using a combined data pool from multiple companies (i.e., customers) to predict security breaches before they occur, and provide a usable set of notifications rather than the long list of "possible alerts" that so many infosec apps drop on users' laps.

Wriber (http://www.wriber.com/) is SaaS that empowers enterprises to build marketing and communications content more effectively using dynamically-generated artificially intelligent writing prompts.

WebAction is working on some complex problems from a data management aspect (continuous queries on streaming, distributed in-memory data). From a front-end dev aspect, we're working on efficient rendering, modular and re-usable UI components and frameworks.

Machine Insight. We have a powerful machine learning system based on genetic algorithms. (http://www.machineinsight.com/employment.html)

We're trying to develop a set of algorithms that can detect if two online accounts are the same person as well as whether or not the person can be trusted.

Not saving the world but its still a fun challenge!

maybe not a hard technical problem; but they're tackling (with science) a gigantic challenge i.e. the global food industry and its discontents.

Hampton Creek http://www.hamptoncreek.com/

This one also looks promising: Endless Mobile https://endlessm.com

and they're also hiring! https://endlessm.com/jobs/

We (Ananas) are working on the challenge of using technology to eliminate extremism amongst other things.

We will be hiring in a few months.

What sort of extremism are you hoping to eliminate??

uBeam: Wireless electricity

Although I dont work there but I think theranos is one such company.

Narrative Science

I had a start up that was working on hard technically challenging problems.

We were a spin-off of the MIT mind machine project. We were working on cognitive modeling with the long-term goal of modeling intention awareness, which is modeling cognition (knowledge and emotions) with a time component.

We were working on knowledge graph theory, blending over time, based on all written communication of an individual or population sub-group.

It proved to be too soon - probably about 10 years too early. We needed more data from individuals - email wasn't enough. We probably needed daily recording of all interactions to build a knowledge graph of an individual. But even working with email data it was too much information - we could barely process all of it especially when blending with a time vector.

Google glass was hopeful for acquiring enough interaction data for a person and their daily lives, but it proved inadequate (no one used it). The closest data set to what we needed was recordings of all interactions on the ISS.

This will be necessary to model a human personality AI at the beta level (copy of an individual based on observations of that person).

We've since moved off on to other projects.

Scrape.it works on building a web scraping tool that minimizes effort and time invested to quickly extract data from websites. Web scraping is simple in theory but building a tool that can work with a lot of different websites in the wild is no easy task.

I am looking for a co-founder.



Could you explain what advance in technology they're working on at Yo? (as opposed to "advances" in social media)



Palantir and yes we are hiring.

How do you justify to yourself working under a CEO who proposed making a smear campaign against Julian Assange?

A third party (HB Gary) — not Palantir and certainly not its CEO — suggested the 'offensive' campaign against Wikileaks.

Palantir — and specifically its CEO — immediately disclaimed involvement or interest in both that plan, or anything of its general type.

How do you justify to yourself slandering a company by claiming nearly the exact opposite of what happened?

That is not true. Palantir along with HBGary and Berico Technologies authored the report suggesting an 'offensive' campaign. [1]

They did apologize for this and as you say, disclaimed involvement or interest in such a plan. [2]

[1] https://wikileaks.org/IMG/pdf/WikiLeaks_Response_v6.pdf

[2] http://www.forbes.com/sites/andygreenberg/2011/02/11/palanti...

The slide deck provides no support for ~adamnemecek's claim that the Palantir CEO himself proposed the plan. And the CEO's unequivocal statements afterwards suggest that if there wasn't a bright-line inside Palantir against offensive operations before 2011, there has been since.

But also, the branding of the presentation does not reliably indicate the origin of the specifically offensive bullet points (most notably the "Potential Proactive Tactics" slide), nor whether the contents ever progressed past trial-balloon status.

In particular, coverage by Greenwald [1] (working from more email thread context for which I can't find a current online source) identifies HB Gary principal Aaron Barr as the verbatim source of some of the most-aggressive anti-Greenwald language.

So I can see that someone at Palantir was coordinating with the other entities and leading the document-prep. But they also cut-and-pasted material from the potential contract partners, including vague proposals outside Palantir's focus, that may never have been approved or presented.

I'd ascribe those bullets to Palantir itself if that deck was in fact ever presented to a client. We don't know if that's the case. If instead that text just bounced around as a discussion draft (including their more-aggro partners), but went nowhere, and was eventually categorically disavowed... then it's a stretch to say that the company or its leadership actually advocated those steps.

[1] http://www.salon.com/2011/02/15/palantir/

Have you read the entire Salon article you linked to?

"So apparently, if Palantir’s new version is to be believed, a 26-year-old engineer went off on his own and — without any supervision or direction — participated in the development of odious smear campaigns intended for two of the nation’s deepest-pocket organizations (Bank of America and the Chamber), potential clients which the emails repeatedly emphasize would be very lucrative. I’ll leave it to others to decide how credible that version is, but I will note that several facts undermine it: ..."

"The leaders at the very top of Palantir were aware of the Team Themis work, though the details of what was being proposed by Barr may well have escaped their notice."

Because it's part funded by the US Govt Intelligence agencies, and it's many clients are also in that sphere.

Nah. Palantir isn't a startup anymore. It's 100% Enterprise, and it "caught" that environment from its (enormous, bureaucratic, government) clients.

All of them

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact