Python AI Developer Needed to Build a RAG-Based Customer Support Assistant

We're looking for an experienced Python AI developer to help us build an AI-powered customer support assistant for our business. We currently have product documentation, FAQs, internal guides and customer support information spread across different documents. Our goal is to make this information accessible through a single AI assistant that can answer customer questions accurately instead of relying on our support team to handle every repetitive inquiry. The first version should be focused on document-based question answering, with the ability to connect to our existing business tools in a later stage. What needs to be built The initial scope includes: Build a RAG-based knowledge assistant that can process PDF, DOCX and other common document formats. Implement document parsing, chunking, embeddings and vector search. Integrate an LLM such as OpenAI or Claude to generate contextual answers. Ensure responses are grounded in the uploaded knowledge base, with references to the original documents. Develop a Python backend using FastAPI. Implement conversation history and basic session management. Add a simple API endpoint that allows our existing application to communicate with the assistant. Include fallback handling when the system cannot find a reliable answer. Provide basic logging and an evaluation process to check answer accuracy. We're open to recommendations on the vector database and retrieval architecture. PostgreSQL with pgvector, Pinecone or a similar solution would be fine, depending on the requirements. We are particularly interested in someone who understands how to make RAG systems reliable in real-world usage, rather than simply connecting an LLM to a vector database. Paid Technical Test Task Before moving into the complete project, we'd like to start with a small paid technical task. We'll provide a few sample documents and a set of questions based on their content. The task will be to: Build a small working RAG prototype using the supplied documents. Implement document retrieval and answer generation. Return source references with each answer. Test the system against approximately 5–10 questions. Briefly explain the retrieval approach and any limitations identified. This will be a paid initial milestone. We're interested in understanding your technical approach, code quality and ability to handle cases where the information is not available in the documents. Engagement We're looking to start with the initial prototype and, if things go well, continue with the full implementation. Please share a couple of examples of RAG or AI knowledge assistant projects you've worked on. When applying, briefly explain: How would you approach building a reliable RAG pipeline for this use case? Which vector database would you recommend and why? How do you handle hallucinations and questions that cannot be answered from the knowledge base? Are you comfortable starting with a paid technical assessment before proceeding to the full project? We're looking for someone who can take ownership of the technical implementation and help us build something that can actually be used by customers. Suggested skills to attach: Python Artificial Intelligence AI Agent Development Retrieval-Augmented Generation LangChain OpenAI API FastAPI Natural Language Processing Vector Database API Integration