All projects
AI & LLM Systems
Ultra-Fast Natural Voice Retrieval-Augmented Generation Engine
Offline RAG system combining vector document search (Phi-3:mini via Ollama) with natural human speech synthesis, delivering grounded audio answers in under 4 seconds.
RAGOllamaPhi-3:miniVoice AIVector SearchStreamlit
The Problem
Why this matters
Traditional document Q&A tools require users to read long text responses. Existing voice RAG solutions suffer from high latency (>15 seconds), making spoken interaction frustrating.
Architecture
System Design
Local Streamlit UI connected to an offline vector index and Ollama running Phi-3:mini. Document chunks are retrieved via semantic embedding comparison, fed to the LLM for grounded answer generation, and synthesized into speech via low-latency local TTS.
AI Intelligence
Machine Learning & Agents
- Document Chunking & Vector Indexing for instant similarity lookup.
- Phi-3:mini LLM for grounded, hallucination-resistant answer generation.
- Sub-4-second speech synthesis engine converting text answers into natural audio.
Stack
Technologies Used
PythonStreamlitOllamaPhi-3:miniVector StoreTTS Engine
Ready to dive deeper?
Explore the architecture, commits, and implementation details of this project.