IRIS — Generative AI Investigation Assistant
Production GenAI assistant for law enforcement and intelligence agencies, built at JSI
IRIS is the generative AI assistant at the core of JSI’s 4Sight investigation platform — a purpose-built intelligence platform used by law enforcement agencies, national security organizations, and defense intelligence teams.
IRIS enables analysts to perform conversational search across large-scale operational datasets, automatically generating context-rich insights about entities, locations, and topics relevant to active investigations. The system is built on an agentic RAG architecture that combines vector search, multi-agent reasoning, and LLM inference to translate natural language queries into actionable intelligence.
4Sight is deployed entirely on-premises with strict access controls and comprehensive audit logging, built to meet current and evolving AI compliance regulations.
My contributions:
- Architected and deployed the agentic RAG pipeline powering IRIS using Agno and LangGraph
- Built high-throughput data ingestion infrastructure with Apache Flink, LanceDB, and ANN indexing (IVF-PQ, HNSW)
- Developed end-to-end LLM evaluation, fine-tuning (LoRA), and deployment pipelines with vLLM inference serving
Tech Stack: Python, Agno, LangGraph, LangChain, LanceDB, Apache Flink, Amazon S3, vLLM, Docker, Kubernetes