Ask HN: AI hackday at work – what shall I work on?

[Tmkly]

I’m a software engineer (typescript/react (and react native)/lua/bash scripting skills). At work this week we’re having an “AI hackday”; we can work or anything we want, as long as it utilises AI in some way (doesn’t have to be work related). An app, proof-of-concept, etc. work will pay for chatgpt or similar API keys (for a limited time). I’m really struggling to come with a unique and fun idea.

Does anyone have any ideas for a useful, fun thing to work on for 24 hours?

[nashashmi]

Something that can scour a confined set of structured and unstructured data and return information when queried based on that data.

Imagine a bunch of word documents containing letters and reports and memos. Now I want to query all of that info and ask for things based on a prompt and which files would be relevant.

edit: Do tell me if you work on this. I came to "ask hn" to literally ask someone to work on this.

[nashashmi → nashashmi]

maybe this will help: https://github.com/adithya-s-k/omniparse

[nashashmi]

One more idea: an information storer and retriever based on a prompt.

  Remember my credit card number: xxx xx x xx x" 

  What visa card numbers do I have

[jimle_uk]

Bit of a controversial opinion (since we are on a programmer's forum) but if you just want to soley focus on the "AI" part and not get bogged down by the code, use a no-code tool like flowise (https://flowiseai.com/) or even better, n8n (https://n8n.io). You will create 100x more successful "showcase-able" AI experiments in the same time it'll take to spin up one from scratch - and guaranteed to have a lot more fun doing so!

Some inspiration here: https://n8n.io/workflows/?categories=AI

I've been creating both experimental and paid AI workflows in n8n since May. Happy to walk you through how to get started.

[throwaway211]

An AI buzzword detector. Like Battlestar Galactica had the Cylon Detector.

It doesn't have to work, like the Cylon Detector didn't. And it could be used to justify what's written by ChatGPT isn't, just like the Cylon Detector was used to show Cylons were not Cylons.

But most importantly, it can justify you keeping a job throughout the AI bubble buzz, even landing a few bonuses, before it all blows over. Just like the Cylon Detector.

[deckiedan]

Ask it to write a sitcom comedy screenplay about a software team asked to do exactly this same situation, but where it goes disastrously wrong. Give enough info to the prompt to make it parody your own company. Make sure it's not offensive.

Then use an AI image generator to generate cartoon pictures for the characters in the screenplay.

Then use an AI voice system to generate all the voices for it.

Then put it all together as a video, present it as a early storyboard concept pitching to them for funding to turn into a full TV series.

[tengbretson]

The knee-jerk anti-hackday reactions are all so tedious. I guess at the end of the day we all get the autonomy we deserve.

[moooo99 → tengbretson]

I completely agree. The hack-day we had at work was the most fun work day I had in the last year. Getting to spend paid working time building something cool, without being tied to any stakeholder input or organization processes was a super fun experience.

[wildrhythms → tengbretson]

A hammer in search of a nail.

[ianbicking]

LLMs are the easiest thing to use, especially if it's all one day. They are also particularly good at data transformation.

Find some data... if nothing else maybe source code or internal docs (if you can't get business data). And think about how you'd want it transformed. Graphviz graphs, CSVs, RSS... some data format that is well understood (including by the LLM), and ideally where you can dump the data into some app or visualization and you don't have to implement that part.

Now your task is collecting and chunking the data, writing the prompts to transform it, and putting the output somewhere interesting (and iterate). Doable in a shortish amount of time but can still be very cool with the right data and prompts.

[spaceywilly]

The best uses of AI I’ve come across are one that are integrated well into a tool I’m already using and just make my job a little bit easier. So look for something where you could add a little box that says “summarize the messages in this thread” or “suggest a title for this note” or something along those lines.

[malux85 → spaceywilly]

* and that work transparently -

Don’t put a button that says “suggest a title for this note” - make the LLM create a title, put it in a placeholder that the user just has to press “TAB” (or something) to confirm

[palata]

I would work on generating phishing emails based on public information I could collect about colleagues/managers. Bonus if you manage to trick the CEO. Call this an "AI-powered security audit".

[f0e4c2f7]

Lately I've been experimenting with using TDD approaches for LLM assisted coding. There is a little bit on YouTube and some research papers on this topic. But most of it relates to how well you understand the test suites for the given language / framework to ask for the tests to be designed in the right way.

This has been great in gpt-4o for python because it has a built in interpreter. So I can describe the way I want the API to look, ask it to write tests in Python, write the code to satisfy those tests, and then run the unit tests in it's own local python env to see if the code works.

This streamlines things a lot. I find I can get working artifacts from gpt this way when other I often have to have a bit of a back and forth where I'm the one testing and pasting in errors.

For gpt-4o this only works with python though (for now).

It might be interesting to build a platform like this, even in your local laptop that allows the LLM to shell out to a sandbox for something like React with the unit tests configured so the LLM can run them and get the output.

Interestingly, frameworks like react also have test suites around ui/ux concepts. I haven't tried it, but I think it would be interesting to use that same approach with those test frameworks to attempt to get better looking frontends (something the LLMs are not especially good at out of the box)

[muzani]

Our winning entry was an automation tool. We'd write down the kind of tests we want to see. It would convert it to cucumber, write the Jira descriptions, then make a git PR, link the ticket, and create the notification on Slack to approve the PR. We already had the systems to run the tests on the PR which was an added bonus.

[iamleppert]

Hackdays are great ways to work with other people in the company and show your product chops to executives. You should ask yourself, what is your strategy for this hackday? Is it to impress an executive, make your boss proud? Don't show too much engagement for the hackday or your boss and executives will question why you don't act like that every day.

[austenvs]

If you have a codebase you work on regularly you could ask AI add sequence comments for Mermaid js to parse into flow diagrams and build the CI tools for it. Maybe it’s separate files with sequence definitions or maybe inline code comment sequences.

[kalaka]

Flunk the office. Have fun outside work.

[teo_zero]

Compare two specifications and summarize the differences (not a word-by-word diff).

[oumua_don17]

I wouldn't invest a lot of mental effort in this as the company will own the idea anyways.

[Tmkly → oumua_don17]

I’m not thinking of anything particularly ambitious. Just something fun I can work on by myself or with a couple of others and can demo to the rest of the company (I work for a pretty small company). They won’t be bothered about “owning” the work.

[purple-leafy]

I’d say if your team have any seeds of really good ideas, don’t work on them at the work hackathon. Don’t work on them at work, period.

Choose the least crap option and work on that.

Work on the great stuff in your own time, and reap your own rewards from it. It’s much harder, but it’s worth it.

[qprofyeh]

Well, if the company is providing input data, it could be interesting to generate something with it. Like a customer profile pic based on their user data and past orders!

[dakiol]

24 working hours or normal hours? If the latter, I wouldn’t recommend doing it. Tired of companies taking advantage of free labour.

[Tmkly → dakiol]

It’ll be normal hours, they won’t expect us to work outside of that.

[aim4min]

Have AI intelligently solve my git merge conflicts. Please.

[add-sub-mul-div]

Ask AI to write you a simulation of a hockey game where everyone skates towards the puck.

[b20000]

are they paying you for the hours you participate in this? if not then you should not participate

[Tmkly → b20000]

It’ll be normal working hours, they won’t expect anything more

[lamroger]

backoffice automation

[meiraleal]

Anything worth it should be done for yourself, not for someone else's company

[cpach → meiraleal]

If OP comes up with three good ideas before the hack day he/she can work on #2 or #3 and keep #1 for themself.

However, some of us have zero desire to launch our own company, so in that case it’s not an issue at all.

[allamaso]

did you asked chatgpt already? lol