Internal knowledge assistant
A retrieval assistant over the company's own documents. Files are chunked and embedded into Weaviate, a question pulls the closest passages, and Claude answers from those passages only, naming the document and section each fact came from.
- Answer time
- about 2 s
- Data
- stays on internal servers
- Interface
- chat and Slack
- Reindex
- automatic on change
Why it exists
Companies have hundreds of internal documents and people still email HR or spend twenty minutes digging through PDFs. New staff can't find anything, so onboarding drags.
How it runs
Ingest
PDFs, Word files and text are split into passages, embedded and stored in Weaviate.
Stay current
When a document changes or a new one is added, the pipeline reindexes it. No manual upkeep.
Ask
Someone asks a plain question in the chat interface or the Slack bot.
Retrieve
The question is embedded and matched against the stored passages. The closest ones are pulled.
Answer
Claude gets the question and those passages and writes a short answer grounded in them only. If nothing relevant was found it says so.
Cite
Every answer names the document and section it came from, so people can check it.
What came out of it
The client reported about 40% fewer internal support tickets after rollout. Answers come back in about two seconds and every one points at its source.
Notes
The "I couldn't find that" path is what makes people trust it. An assistant that guesses when the documents don't say gets ignored within a week. Everything runs on the company's own infrastructure, so no document leaves the building.
Stack
- Python (FastAPI)
- Weaviate
- LangChain
- Claude API
- Docker
- AWS EC2
Want something like this built?
Tell me what you're trying to automate. Most things can be built in a week or two.