Inspiration

Inspired by how perplexity makes our research faster, and how far can this be pushed? Can this be taken to private data sources like telegram channels.

Inspired by how AutoGPT breaks down a goal into subtasks and performs them sequentially. Inspired by how MemGPT incrementally refines the answer.

What it does

Enter any query and autosearch.lol breaks it down into subtasks, performs web and telegram vector databse seraches, then incrementally refines the answer until it comes to the best answer it can.

How we built it

Used Vannever Labs telegram data set, which is then vectorized in the supabase vector db. Then it is searched by the query agents. The agents also search Exa.

Challenges we ran into

Missing data in vannever data set

Accomplishments that we're proud of

It is live at autosearch.lol

What we learned

Parsing csv files.

What's next for autosearch.lol

Improve the prompting and do some marketing online.

Built With

  • llama3
  • nextjs
  • vannever
Share this project:

Updates