Project Overview We are building Yuktha, an AI-driven women’s metabolic health platform (starting with PCOS). Our system uses a Retrieval-Augmented Generation (RAG) pipeline to deliver personalized recommendations (diet, supplements, lifestyle, coaching) via mobile app and WhatsApp. A baseline RAG system is already developed. We are looking for an expert to audit, optimize, and scale the system for production-grade performance. --- Scope of Work 1. RAG System Audit - Review current architecture (retrieval, embeddings, prompting, orchestration) - Identify: - Hallucination points - Retrieval failures - Latency bottlenecks - Context leakage / irrelevant responses --- 2. Retrieval Optimization - Improve: - Chunking strategy - Embedding selection - Query rewriting / expansion - Optimize vector search (recall vs precision tradeoff) - Implement hybrid retrieval if needed (semantic + keyword) --- 3. Prompt Engineering & Response Quality - Redesign prompts for: - Clinical-style accuracy (PCOS domain) - Structured outputs (plans, recommendations) - Reduce hallucinations - Ensure consistency across sessions --- 4. Personalization Layer - Improve user-context handling: - Symptoms - Test results - History - Implement memory-aware responses --- 5. Evaluation Framework - Build evaluation metrics: - Answer accuracy - Relevance - Safety - Create automated + manual evaluation pipeline --- 6. Integration Support - Ensure smooth integration with: - Mobile app - WhatsApp workflows (via APIs) - Optimize response latency (<2–3 seconds target) --- Expected Deliverables - Detailed audit report (issues + recommendations) - Improved RAG pipeline (code + architecture) - Prompt library (modular + reusable) - Evaluation dashboard / framework - Documentation for internal team --- Required Skills Must-Have - Strong experience with RAG systems in production - Hands-on with: - or - Vector databases ( / ) - Experience with LLM APIs ( / ) - Prompt engineering for structured outputs - Debugging hallucinations and retrieval errors --- Good to Have - Experience in healthcare / wellness AI - Knowledge of knowledge graphs - Experience with multilingual systems (Indian languages) - WhatsApp / conversational AI integrations --- Engagement Model - Duration: 4–8 weeks (initial engagement) - Mode: Remote - Commitment: 20–40 hours/week - Potential for long-term engagement --- Selection Criteria - Demonstrated work in real RAG systems (not demos) - Ability to explain trade-offs (precision vs recall, cost vs latency) - Strong debugging and system thinking skills --- How to Apply Please share: 1. Relevant RAG projects (GitHub / case studies) 2. Your approach to improving an existing RAG system 3. Tech stack familiarity 4. Availability and expected compensation --- What Success Looks Like - Significant reduction in hallucinations - Improved relevance and personalization - Faster response times - Production-ready, scalable system --- Note: This is not a basic chatbot project. We are building a high-trust health AI system, and accuracy + reliability are critical.