Afraz K.

I'm a Senior AI Engineer specializing in Generative AI, AI Agents, RAG, Document AI, and OCR. I design and deploy production AI systems using Python, LangGraph, LangChain, LLMs, and FastAPI helping healthcare, fintech, and SaaS teams automate complex workflows, extract reliable data, and turn company knowledge into usable AI applications. I work beyond the prototype: agent orchestration, document ingestion, retrieval, validation, APIs, databases, observability, and deployment. WHAT I BUILD **AI AGENTS & MULTI-AGENT SYSTEMS** - Custom AI agents, agentic AI workflows, and task-specific multi-agent orchestration. - LangGraph / LangChain, ReAct agents, LLM tool calling, MCP integrations, memory, structured outputs, retries, and human-in-the-loop approvals. - Code-first AI automation with Python and FastAPI, connecting agents to business applications, databases, and internal APIs. **GENERATIVE AI, LLM ENGINEERING & RAG** - Production LLM applications and enterprise AI assistants using OpenAI, Claude, Gemini, and open-source models. - Retrieval-Augmented Generation (RAG), GraphRAG, semantic search, hybrid search, vector databases, embeddings, BM25, and reranking. - Knowledge-graph and vector retrieval with Neo4j, PostgreSQL/pgvector, MongoDB, and document repositories. - Citation-grounded answers, evaluation, access controls, and safeguards to reduce unsupported responses. **DOCUMENT AI, OCR & INTELLIGENT DOCUMENT PROCESSING** - OCR expert services and intelligent document processing (IDP) for scanned PDFs, tax documents, invoices, financial statements, ID documents, and forms. - PaddleOCR, Tesseract OCR, AWS Textract, OpenCV, YOLOv8, layout detection, table recognition, and computer vision. - Document classification, field extraction, MRZ parsing, structured JSON, field-level confidence, validation rules, and human-review routing. - Integrations that send extracted data into existing workflows rather than leaving it in a demo interface. **AI AUTOMATION & PRODUCTION BACKENDS** - Custom AI workflow automation, API integration, REST/gRPC services, Python, FastAPI, Docker, PostgreSQL, and Redis. - Architecture for security-sensitive data, reliability, maintainability, monitoring, and practical deployment. **SELECTED PROJECT EXPERIENCE** **1. Healthcare AI Agents:** Built a LangGraph-based ReAct orchestrator connecting clinical documentation, insurance eligibility, medical-record retrieval, and routing workflows through a conversational interface. **2. Enterprise GraphRAG:** Built hybrid retrieval combining Neo4j knowledge-graph relationships and vector search to support more context-aware answers over business data. **3. Financial Document AI:** Developed OCR and layout-aware extraction for complex PDFs, financial tables, and tax-related documents. My Upwork work history includes a paid tax-statement OCR pipeline, table recognition, and PDF detection engagements. **4. Fintech KYC:** Developed identity-document OCR, MRZ extraction, face matching, and validation for onboarding and verification workflows. **5. Healthcare Integrations:** Worked on X12 270/271 eligibility automation, clearinghouse communications, and clinical document processing integrated with backend workflows. **WHY WORK WITH ME** - Top Rated Plus on Upwork with a 100% Job Success Score. - Ongoing Senior AI / Machine Learning Engineer contract and repeat clients for specialized OCR and document processing. - Experience connecting AI capabilities to real systems—not just building chatbot demos. - Clear focus on accuracy, validation, maintainable architecture, and knowing when a human needs to review an AI output. **TECHNOLOGY STACK** Agents & LLMs: LangGraph, LangChain, OpenAI, Claude, Gemini, MCP, tool calling. Retrieval: RAG, GraphRAG, Neo4j, pgvector, MongoDB, embeddings, BM25, reranking. Document AI: PaddleOCR, Tesseract, AWS Textract, YOLOv8, OpenCV, computer vision. Backend: Python, FastAPI, PostgreSQL, Redis, Docker, REST APIs, gRPC. **LET'S DISCUSS YOUR PROJECT** Send me your workflow, system requirements, or a few redacted sample documents. I'll help identify the right architecture, likely failure points, and a practical path from prototype to production.