Labs experiment
Knowledge assistant: answers from company documents, with every claim cited
An assistant that answers questions from a company's own documents and cites the passage behind every claim. Each question is rewritten into a better search, matched against the embedded documents, reranked by relevance and answered with citations.
Where this pattern fits
- Support and operations teams answering the same policy questions
- Onboarding, where new people need answers from internal documents
- Any knowledge base where a wrong answer costs trust
- ChromaDB
- Embeddings
- LangChain
- Next.js
- Claude
How it works
Select any node to see what it does.
- 00.00userwhat is our refund window for annual plans?
- 00.08rewritequery: refund policy, annual subscription
- 00.31search6 chunks from 2 policy PDFs
- 00.92writeranswer drafted with 2 citations
The architecture, step by step
Each part of the diagram, in the order a run moves through it.
- QuestionWhere every run starts.
- RetrieverFinds the passages that answer the question.
- Query rewriteRewrites the question into a better search.
- Vector searchSearches the embedded documents.
- RerankerOrders the passages by relevance.
- Answer writerWrites the answer with a citation for every claim.
- ToolsThe agents act through the ChromaDB, PDF parser and citations.
How it is built
The design choices that make the pattern hold up with real users.
- A citation on every claim, so each answer can be checked against its source
- Query rewriting and reranking for better recall than plain vector search
- A PDF parser that keeps the source of every chunk in ChromaDB
- LangChain for the pipeline and Claude for the answers, behind a Next.js interface
Related
Pathfinder
A multi-agent assistant for the whole job search.
Study Mode
Gemini-powered study sessions built from your own notes.
AI Integration
Add a real LLM feature to the product you already have. From $1,000, 2 to 3 weeks.
Voice receptionist
An AI phone agent that answers, books and hands off to a human.
CRM tool server
A Model Context Protocol server that gives Claude safe tools for a CRM.
Evals and tracing
Every agent run traced, scored and compared before it ships.