A voice-first Retrieval-Augmented Generation system designed with strict zero-trust data controls. Allows enterprise operators to query massive unstructured document archives via real-time speech while preventing data exfiltration.

Tested and benchmarked against the 10,000+ email Enron corpus across 77,525 document chunks. Incorporates automated PII detection and redaction, dynamic context chunking, Web Speech API speech-to-text with typed fallback, and a high-throughput Python FastAPI backend.