AWS RAG Voice Bot Build

I’m building a production-ready voice assistant that uses a large language model reinforced by Retrieval-Augmented Generation (RAG) to give accurate, context-aware answers in real time. The entire stack must live inside my AWS account so I stay in control of costs, data, and scaling. Here’s the core of what I need • A conversation engine that takes spoken input, runs it through speech-to-text, passes the transcript to an LLM, and returns a spoken response (text-to-speech). • A RAG layer that can call custom knowledge sources—think S3 documents, DynamoDB records, or an Aurora cluster—via vector search / embeddings so answers are grounded in my own data rather than generic web knowledge. • Event-driven automation hooks (e.g., trigger a Lambda, post to an SNS topic, or call an API endpoint) whenever certain intents or entity combinations are detected. • Clean AWS deployment: infrastructure-as-code (CloudFormation or Terraform), clear network boundaries, logging, and basic CI/CD so I can replicate the stack in another account with a single command. Deliverables 1. Source code and IaC templates stored in my private Git repo. 2. Step-by-step deployment guide and short screencast walk-through. 3. A sample knowledge base plus at least five end-to-end conversation test cases proving the RAG layer is actually being hit. 4. A quick performance report (latency, cost per call) from a small load test. Open to your preferred LLM, vector store, and AWS services as long as they fit within the account and can scale. If you’ve shipped something similar, point me to a demo or outline your approach in your bid.

Project ID: 40210319