Hi everyone, it's Alessandro here, and this is my first post on HN wuhuuuu :D
Anyways, I'm writing here because I wanted to get your constructive feedback on a project I recently developed, GreenReguAI (website: https://greenreguai.com/, video presentation: https://www.youtube.com/watch?v=0CTiVbwfIfo).
To give you a bit of context, I've got a background in Energy Engineering, and I worked a couple of years in a German company that builds large-scale photovoltaic and storage plants around the world. During my time at this firm, I was always baffled by how inefficient the process of regulatory compliance was. Basically, when you need to build a plant of this scale somewhere, you need to check loads of local laws to ensure that the project you want to build respects all of them.
This shouldn't be too hard to do theoretically, because these kinds of plants have been built for a long time, so the process should be pretty standardized, right?
Well, it should, but it's not that easy, at least not in Europe (and I suppose also in other parts of the world). It turns out that despite being one of the continents more favorable to renewables, Europe has quite fragmented legislation, and when one wants to build such a plant, they get lost in a jungle of ugly, hard-to-navigate websites that seem to have been developed in the 00s and never updated, multiple sources of truth with answers hidden in a small paragraph contained in a document of hundreds of pages, and all of this in a context that changes with a non-negligible frequency, because regulation evolves over time.
I know this because once I had to make such an analysis for a storage plant in Poland, and it sucked.
Therefore, I thought that with the new technological development, this issue could, if not be completely solved, at least be made better.
GreenReguAI is atm a RAG connected to a commercial LLM and a database where many documents relevant to this context are stored, processed, divided into chunks, and embedded. The differentiating factor is that for each chunk I save a few metadata, including which document, page, and position in the page the chunk comes from. This enables the creation of a quite detailed 'source verification' feature that, in 1-click, enables the user to verify where each part of the answer provided by the AI Assistant comes from, even accessing the exact position in the original document. This feature is not like the ones of Perplexity, ChatGPT, or other similar products, which in their answers reference a certain source, but that's it, you need to navigate the source by yourself to find where exactly the info on which the answer is based is contained. In GreenReguAI, when you click on the source of the answer, you get redirected to the exact place where the info is, which is something essential when you need to work with super boring legal documents that are sooooo long.
If there is enough interest in this product, I'm already planning some improvements:
1) Make the RAG more 'agentic,' so that instead of a one-shot approach, the RAG agent can perform multiple queries and research, which should improve the quality of the answers.
2) Add a 'Sources Scouting Agent' that is connected to a few websites that publish the official sources we are looking for (like websites of governments, international agencies, and legislative entities) and automatically downloads the new ones relevant to this context once they are made available.
This is overall the idea, I would love to hear what you think about it <3
