Skip to content

Latest commit

 

History

History
59 lines (39 loc) · 2.63 KB

File metadata and controls

59 lines (39 loc) · 2.63 KB

knowledge-navigator

A RAG agent that answers questions against any document corpus with citations.

CI Phase Python

Problem Statement

Organizations with large document libraries (quality manuals, standards, procedures) spend significant time manually searching for policy answers. This agent provides instant, cited answers to natural-language questions — reducing lookup time and ensuring answers are traceable to source documents.

Architecture

Architecture diagram added Week 4 of Phase 1. See ARCHITECTURE.md for component breakdown.

Agentic Pattern: RAG (Retrieval-Augmented Generation)

Rather than relying on an LLM's training data (which may be outdated or hallucinated), RAG agents retrieve relevant document chunks first, then generate answers grounded in those retrieved sources. Every answer includes citations showing exactly which document sections were used.

Evaluation Targets

Metric Target Method
Retrieval Accuracy ≥85% 25 curated Q&A pairs
Citation Correctness ≥90% 50 spot-check citations
Response Time <8s Automated timing

Quick Start

Active development begins Week 4 (Phase 1). Full demo available at Phase 1 completion.

git clone https://github.com/Jlowpez/knowledge-navigator
cd knowledge-navigator
pip install uv && uv pip install -e ".[dev]"
pytest tests/ -v

Production Path

Replace the document corpus in data/synthetic/ with your organization's documents. The ingestion pipeline and agent logic are unchanged — only the source files differ. The data access layer (src/dal/) abstracts all backend calls, enabling a data source swap via config change only.

Part of a Larger System

ops-orchestrator — full platform view

Agent Pattern Status
knowledge-navigator RAG Agent 🔄 Active
intake-agent Structured Output ⏳ Phase 2
compliance-checker Multi-Agent Pipeline ⏳ Phase 3
workload-scheduler Planner-Executor ⏳ Phase 4
ops-orchestrator Orchestrator ⏳ Phase 5

See Also