No More Hallucinated Citations: A Domain-Specific RAG System with Ollama, ChromaDB and AI Agents

Short summary

Developer built a domain-specific RAG system combining ChromaDB, Ollama, and Claude Code agents to eliminate hallucinated citations from academic research. Uses hybrid relevance scoring on a corpus of 2,514 multilingual PDFs with 59,030 indexed chunks. System runs locally with zero-hallucination citations and evidence-backed research output through FastAPI endpoints and AI agent slash commands.

• •Self-hosted RAG pipeline for 2,514 academic PDFs in 4 languages with 99.8% coverage
• •Hybrid relevance scoring combines vector similarity, Ollama quality scores, and publication recency
• •Claude Code agents orchestrate research workflows with verified citations and traceable sources

Generated with AI, which can make mistakes.