LLM-Assisted OpenSearch

This project addresses a critical challenge in criminal investigations: the difficulty of searching through vast volumes of unstructured evidence while bridging the semantic gap between natural language investigative intent and technical search logic.

The project develops a cloud-native microservices architecture specifically tailored for private cloud deployments in investigative settings. At its core, the system integrates Large Language Models (LLMs) into a “Human-in-Control” workflow that translates natural language queries into syntactically valid OpenSearch Domain-Specific Language expressions, ensuring investigators maintain oversight and control throughout the search process.

A key innovation is the implementation of a hybrid retrieval strategy within OpenSearch that combines BM25-based lexical search with nested semantic vector embeddings. This dual approach leverages both traditional keyword matching and modern semantic understanding to improve search accuracy and relevance.

The project delivers a functional prototype demonstrating technical feasibility, establishing an architectural baseline for future empirical evaluation. A rigorous evaluation methodology is outlined using the Enron email dataset as a structural proxy for restricted investigative corpora, enabling validation without compromising sensitive law enforcement data.

This work provides a foundation for enhancing investigative capabilities through AI-assisted search while maintaining the essential human oversight required in law enforcement contexts.