· Serverless Melbourne meetup
A generative AI second brain on AWS serverless
How to keep a personal knowledge base on AWS without paying for a large model on every document.
A "second brain" is a place outside your head where notes, documents and ideas are kept so you can find and use them later. The talk is mostly about keeping one cheap to run.
Most of the work a second brain does doesn't need a large model or an instant answer, and a few mistakes are fine. So documents are stored in S3, with Glacier for old ones, and Step Functions runs the steps. Small BERT-based models pull out keywords and topics to organise them, and LanceDB keeps a full-text and vector index on S3. Small quantized models run in Lambda or on spot instances for the cheap steps. Retrieval with a larger model runs only on content I pick.
What it covers
- What a second brain is, and the capture, organise, distil and express method
- Why serverless suits it, and what the managed AI services would cost
- Which open-source tools do the cheap parts
- The architecture that ties it together