I want to to deploy a production-grade, retrieval-augmented (RAG) chat agent that can confidently answer customer inquiries with accurate product details drawn from my existing knowledge base. The data is already in-house, well structured, and waiting to be piped into the retrieval layer—you won’t need to source or label anything new, just design the ingestion and indexing workflow so it scales. We’ll work shoulder-to-shoulder during key milestones; I’d like live screen-share sessions to review the architecture, prompt design, vector store choices, and integration points. Think of it as pair-building: you lead the implementation while I supply domain context and immediate feedback, keeping iteration tight and decisions transparent. Key deliverables • End-to-end RAG pipeline (data ingestion, vector indexing, retrieval, generation) • Chat interface or API endpoint ready for production traffic • Deployment scripts/infrastructure as code (Docker/Kubernetes or equivalent) • Brief run-book and hand-off session so my team can monitor, retrain, and extend the model Acceptance criteria 1. Answers must reference source passages returned by the retriever. 2. Latency ≤ 2 seconds on a 20-document context window. 3. At least 90 % accuracy on a validation set we’ll craft together from real support tickets. Preferred stack includes Python, LangChain or LlamaIndex, OpenAI or similar LLM endpoints, and a vector database such as Pinecone, Milvus, or FAISS—but I’m flexible if you can justify alternatives. We need to test multiple LLM models, and use the mix of cost effective If you have shipped a RAG or similar conversational AI to production and enjoy collaborative build sessions, let’s get started. -- 1️⃣ AI User Support Agent (Rules, Formats, Pricing, How-To) Purpose: Reduce support load (calls/emails) and help users instantly understand CricBattle, formats, rules, and pricing. Responsibilities: • Answer “What is CricBattle?” and “How does it work?” questions • Explain all fantasy formats (Auction, Draft, Salary Cap, Run Leagues, Prediction, etc.) • Explain rules in simple language with examples • Clarify pricing models (free vs paid customization, CB+, corporate pricing) • Explain differences vs Dream11 / DFS (season-long vs daily) • Guide new users step-by-step (“How do I create a league?”) • Handle FAQs like join issues, deadlines, scoring confusion • Suggest relevant help articles or follow-up questions • Detect confusion / frustration and simplify responses • Escalate complex issues to human support (flag only, no resolution) Notes: • Must use knowledge-base driven Q&A (RAG) • Must never hallucinate rules not supported by CricBattle • Tone: friendly, simple, non-technical