AI Powered Document Platform

The problem

Organisations sit on large volumes of documents that contain valuable knowledge but are difficult to search and interrogate. Traditional keyword search misses context and requires users to know what they’re looking for.

The solution

A full stack platform with a RAG powered backend for intelligent document interaction. Users upload documents which are processed through an ingestion pipeline (chunking, vector embeddings via text embedding 3 small, pgvector storage), then query them in natural language through streaming chat responses.

Key features

• Document ingestion pipeline with chunking and vector embeddings
• Streaming chat responses with an LLM agent and conversation history
• Semantic search across uploaded documents
• Ephemeral document sharing with configurable session expiry and query quotas
• Partner matching

Testing and architecture

Nullable infrastructure wrappers for database, storage, and HTTP layers with configurable responses and output tracking, enforced by architectural tests (pytestarch). Zero mocking frameworks.