Docsight
Internal AI assistant that answers employee questions from company documents, with source references and access control.
Concept / Internal Project
This is a concept project created to show how we approach this type of product. It is not a client engagement, and it contains no client data, metrics or results.
The problem
Company knowledge is spread across documents, wikis and drives. Employees spend time searching or asking colleagues, and generic chatbots can't see internal data — or see too much.
Proposed solution
A retrieval-augmented assistant: documents are indexed with their access rules, answers always cite sources, and quality is checked against a test set before each release.
Architecture
Ingestion
Python workers that parse documents, split them into chunks and create embeddings.
Retrieval
PostgreSQL with pgvector; results are filtered by the user's permissions.
Answering
LLM API with prompts that require citations and allow "I don't know".
Interface
Next.js chat UI with source previews and feedback on each answer.
Start here
Building something similar?
Tell us about your product and we'll outline how we'd approach it.
What happens next
- 1
We review your request
And come back with clarifying questions.
- 2
Intro call
We discuss goals, users, constraints and timing.
- 3
Scope & estimate
You get a proposed scope, team, timeline and budget.