Project Mockup
Current Scope
Personal AI Health Assistant
We are building a comprehensive Personal AI Health Assistant Agent that empowers users with instant, intelligent health guidance across four specialised AI agents: Doctor AI for symptom checking and urgency detection, Fitness AI for personalized workout plans and exercise form analysis, Nutrition AI for meal planning and calorie estimation from food photos, and Converse AI for general wellness conversations. The platform is powered by Gemma 4 and Qwen LLMs with a full RAG (Retrieval-Augmented Generation) system that grounds responses in a curated medical and fitness knowledge base — minimising hallucinations and ensuring accuracy. Built with bilingual support (English / বাংলা), voice input, chat memory and history, multi-account management, query suggestions, source citation, and a real-time admin dashboard with analytics and feedback tracking.
Tech Stack
Challenge & Solution
Challenge
- Ensuring medical accuracy with LLM hallucination risk
- Bilingual NLP for English and Bengali
- Real-time voice-to-text integration
- RAG pipeline with domain-specific knowledge base
- Multi-agent context isolation and memory management
Solution
- RAG system grounding responses in verified medical/fitness knowledge
- Gemma 4 & Qwen LLM integration with fine-tuned prompts
- Bilingual tokenization and language detection pipeline
- WebSocket-based voice streaming with Web Speech API
- Per-user conversation memory with PostgreSQL session store