VoiceRx
VoiceRx is an AI-powered voice assistant designed to provide accessible preliminary health information to patients who lack immediate access to medical professionals.
The Problem
In many regions, wait times for doctors are extensive, and patients lack a reliable way to get immediate, conversational answers about symptoms or medications without resorting to anxiety-inducing web searches.
The Solution
A voice-first interface powered by Vapi and Groq, utilizing a custom RAG (Retrieval-Augmented Generation) pipeline backed by Qdrant vector database to ensure medical accuracy.
Architecture
FastAPI backend handles webhooks and RAG logic. Qdrant stores medical embeddings. Groq provides ultra-fast LLM inference, and Vapi orchestrates the voice synthesis and recognition.
Engineering Challenges
Reducing latency in the voice pipeline. The combination of STT (Speech-to-Text), RAG retrieval, LLM generation, and TTS (Text-to-Speech) initially took over 2 seconds. Optimized by streaming chunks and using Groq.
Results & Impact
Selected as a Finalist at HackBLR 2026 out of 2,500+ participants across India.
Lessons Learned
Deep dive into real-time voice architectures and latency optimization for LLMs.