← Back to projects
Knowledge
Enterprise RAG Platform
Document-intelligence platform with intent-driven retrieval, a feedback-driven evaluation harness, and citation-grounded answers that refuse unsupported claims.
How it works
A Vue frontend streams chat and admin workflows from a FastAPI backend. LangGraph-style workflows coordinate ingestion, strategy-aware retrieval, citation validation, and evaluation, persisting state in SQLite and vectors in Milvus.
Vue 3FastAPILangGraphMilvusSSEDocker Compose
Capabilities
- Intent-driven retrieval strategies (precise, balanced, broad) over hybrid search, query expansion, HyDE, and RRF reranking.
- Citation-grounded streaming answers with source validation and strict-evidence refusal.
- Feedback-driven golden set: user thumbs-up/down promoted into evaluation cases with expected chunks.
- Retrieval, answer, and judge-based regression checks with per-strategy breakdown.
- Real auth with entity-level access control over documents, citations, and sources.
Evidence
- Inspectable chunks, retrieval traces, and citation metadata — a test bench that debugs retrieval without spending LLM tokens.
- Golden-set evaluation for retrieval-only, answer-lite, and full judging.
- Strict-evidence mode refuses unsupported facts.
Try this
- Look up an exact policy amount with precise retrieval.
- Ask a broad question in broad mode.
- Enable strict evidence and ask for an unsupported fact to see refusal behavior.