Doclantern — document search assistant
A concept assistant that answers questions from a company's own documents: staff ask in plain language and get an answer with the source passages it came from.
This is an illustrative concept study, not a client engagement. It describes a reference approach rather than delivered results.
Company knowledge lives in policies, handbooks and project documents that nobody can search effectively. Staff either interrupt a colleague or give up — and both cost time.
The challenge
Company knowledge lives in policies, handbooks and project documents that nobody can search effectively. Staff either interrupt a colleague or give up — and both cost time.
Our approach
Doclantern uses retrieval-augmented generation: questions are matched against the company's indexed documents, and the assistant answers only from what it found, citing the source passages. When the documents do not contain an answer, it says so.
Key features
- Plain-language questions answered from indexed documents
- Source citations with every answer
- Honest 'not found' responses instead of guesses
- Admin controls for which collections are searchable
The stack
A managed vector index over the document library, a server-side AI gateway for the model, and React for the chat interface.
What this demonstrates
A grounded AI pattern: useful answers with visible sources, and no invented facts when the documents are silent.
Related projects
Let's talk
Have an idea? Let's figure out how to build it.
You don't need a technical specification or a perfectly defined plan. Tell us what you're trying to achieve, and we'll help identify the next steps.