I will build an AI assistant over your documents with Claude
About this service
A retrieval-augmented assistant that answers from your own content and tells you when it does not know. I build the ingestion pipeline, the retrieval layer, the evaluation harness and the guardrails — in that order, because a RAG system without evaluation is a demo, not a product.
Built on the Claude API with pgvector or your existing vector store.
Skills
What the seller needs from you
You will answer these right after ordering — the delivery clock starts once you do.
- 1What kind of documents, and roughly how many?required
- 2Who will use the assistant and what will they ask it?required
- 3Do you have a preferred model provider or hosting constraint?
Frequently asked
Where does my data live?
Wherever you want it to. Default is your own Postgres with pgvector; nothing is sent anywhere except the model provider you choose.
How do you measure quality?
A held-out question set with graded answers, scored before and after every retrieval change. You get the harness, not just the numbers.
