Senior AI Engineer (Python, RAG, LangGraph, LLM, Agents) Needed

We're looking for a skilled AI Engineer to build an AI-powered knowledge retrieval solution using RAG (Retrieval-Augmented Generation). The solution will process PDF, DOCX and XLSX documents and allow users to ask questions through a conversational AI interface, receiving contextual answers with accurate, page-level source citations. Initially, this will be a standalone, Dockerized application with a FastAPI backend and React chat UI. It will later integrate into our client’s existing CRM (Interior company) as an independent microservice. WHAT YOU'LL BUILD 1. DOCUMENT PROCESSING - Document upload, listing, deletion and reprocessing. - Text extraction from PDF, DOCX and XLSX. - OCR for scanned PDFs and table extraction where required. - Metadata extraction, document chunking and page/sheet tracking. - Asynchronous processing with status tracking. 2. RAG & AI PIPELINE - Embedding generation and vector indexing. - Semantic search and metadata-based retrieval. - Hybrid search and reranking where appropriate. - Context-aware question answering using LLMs. - Conversational follow-up questions and query rewriting. - Safe fallback when information is unavailable. - Basic evaluation using a test question set. 3. LANGGRAPH ORCHESTRATION - Implement a simple, modular and explainable RAG workflow using LangGraph. - Orchestrate retrieval, context preparation and answer generation. - Maintain conversational context and state. We prefer purposeful LangGraph implementation over unnecessary multi-agent complexity. 4. CITATION ACCURACY (MANDATORY) - Answers must include source document names, page numbers and relevant excerpts. - Citations must map to actual extracted document content and original page locations. - Page references must be validated, not generated or guessed by the LLM. - Provide a clear fallback when supporting information is unavailable. 5. CHAT UI & APIs - Simple React-based chat interface. - Document management and conversation history. - Clickable source references. - FastAPI backend with documented REST APIs. - Docker-based deployment. TECHNOLOGY STACK Python, FastAPI, LangGraph, LangChain/LlamaIndex, OpenAI, Claude, Gemini, pgvector (preferred) / Pinecone / Qdrant / Weaviate, Cohere or BGE rerankers, Docling / Unstructured / Tesseract / PaddleOCR / Textract, RAGAS / Langfuse, Celery/Redis, React, Docker. We are open to technology recommendations. The freelancer should explain their choices based on scalability, performance, maintainability and cost. WHAT WE WILL PROVIDE - 10 to 20 actual project documents, including scanned documents where available. - Sample PDFs, Word and Excel files. - Example questions from the sales team. - Business context and LLM API access/budget. - CRM architecture details for future integration. - Feedback and testing support. DELIVERABLES - Complete standalone solution. - Functional RAG and LangGraph implementation. - Conversational chat UI. - Accurate source citations and page references. - FastAPI APIs and Docker deployment. - Complete source code and technical documentation. MANDATORY DEMONSTRATION The freelancer must demonstrate the working application using our sample documents, covering document processing, multi-document Q&A, follow-up questions, citation accuracy and unavailable-answer handling. FUTURE ROADMAP The service will later be integrated into our existing CRM. Future opportunities include customer-specific retrieval, similar project recommendations, historical estimation analysis and AI-assisted proposal generation. WHO WE'RE LOOKING FOR - Strong Python and FastAPI skills. - Practical RAG and LLM application experience. - Understanding of embeddings, vector databases and semantic search. - Experience with LangGraph or similar orchestration frameworks. - Knowledge of document processing and OCR. - Focus on answer quality, citation accuracy and maintainable code. New freelancers are welcome if they can demonstrate strong technical capability. TIMELINE & ENGAGEMENT Target: 1 to 2 weeks for the initial working MVP, depending on document complexity. This is intended as a long-term engagement, with opportunities for CRM integration, optimization and further AI development. IN YOUR PROPOSAL Please include: 1. Relevant RAG/LLM projects you have built. 2. Proposed architecture for document processing, retrieval and vector storage. 3. Your approach to LangGraph orchestration. 4. How you will ensure accurate page-level citations. 5. Estimated timeline, milestones and fixed-price cost. Start your proposal with: KK-4827 Also briefly answer: - How would you process scanned PDFs containing tables? - How would you prevent answers from citing incorrect document pages?