Irfan U.

I help startups and product teams turn their documents and data into reliable AI & RAG systems that answer with sources, document AI that extracts data you can audit, and AI agents that act with human approval. Top Rated, 100% Job Success, PhD researcher in AI, 10+ years of software engineering. WHAT I BUILD - RAG chatbots and knowledge bases over your PDFs, docs and databases. Hybrid and vector search (pgvector, Qdrant, Pinecone), reranking, citations on every answer, knowledge graphs (Neo4j/PostgreSQL). - Document AI and OCR. Extract fields, entities, tables, claims and their sources from scanned PDFs, forms, invoices, customs and legal documents into structured data, with confidence scores and human review. - AI agents. Tool-calling agents (LangGraph, OpenAI, Claude, MCP) connected to your APIs, with approval steps before anything important happens. Every system ships with evaluation (test sets, retrieval metrics, hallucination checks, regression tests) and production deployment (FastAPI, Docker, AWS/GCP, CI/CD, monitoring), plus React/Next.js screens when your team needs them. RECENT WORK - Customs declaration automation (Australia): OCR + LLM extraction from shipping documents into declarations. [time saved / accuracy / volume] - Explainable CV-to-job matching service: vector retrieval plus transparent scoring. [scale / latency] - AI matching and multi-posting for a job board platform (5-month contract). [result] - Scoring engine and data pipeline integrated into a client platform. [result] - Mood-based NLP search and recommendations for places. [result] HOW I WORK I start with a short discovery: your data, what "correct" means and how we'll measure it. Then milestone delivery with a demo and short written update at each step. You get code, documentation and an evaluation report you can rerun. Have documents, data or a workflow you want AI to handle reliably? Send me a message with a sample and I'll suggest an approach.