albertdessaint

Hi HN, I’m building a chat app where users can create group chats that include an AI. The idea is that group members can ask questions or share information with the AI, and the AI should be able to remember who said what, for better context and retrieval later on.

I’m exploring RAG-based approaches and looking into GraphRAG (like CircleMind or Neo4j-based solutions), but these seem complex and potentially expensive to scale.

Do you think GraphRAG is the right fit for this kind of group-context memory? Or would a simpler architecture be more appropriate? Any advice or alternatives would be much appreciated!

Thanks a lot!

LLMs thrive on high-quality data. While features like web search, file uploads, and integrations with diverse data sources are useful, there seems to be no system for sharing and accessing information between LLM users.

Take my case as an example: I lend money to businesses. Ideally, if someone asks ChatGPT a question like "How can I get B2B financing?" I’d want them to discover me and then be able to ask follow-up questions about financing options, eligibility criteria, required documents, and the process. From there, I could step in and provide personalized assistance.

While web search exists, the internet wasn’t built for the LLM era, and appearing in search results isn’t guaranteed. I’ve also thought about creating an AI agent, but discoverability remains the issue.

Wouldn't a new information system tailored for LLMs be valuable? Imagine an open ecosystem where structured and unstructured data can be shared and accessed by LLMs on behalf of users, powering apps and tools that unlock opportunities and save time.

Am I missing something here, or does this idea have merit?