Azib Z.

AI Engineer | RAG, LLM Agents, Automation | LangChain + n8n | US

AI/ML Engineer — RAG, LLM agents, AI automation, and production MLOps. MS in AI/ML (University of Pittsburgh), US-based, 100% Job Success. 93% retrieval recall, 78% less wasted context, 35% faster hiring. Most AI projects die at the demo. The notebook works, everyone claps, and nobody can ship it. I build LLM and ML systems that survive production. Send me an invite and you'll have a reply within the hour — plus a short Loom walking through how I'd build it, roughly what it costs, and whether it's worth building at all. No obligation, and if the answer is "don't build this," I'll say so. WHAT I'VE SHIPPED — JoinEight: AI hiring engine that cut candidate shortlisting time 35% and lifted offer acceptance — Chex: computer-vision vehicle inspection that sped up checks 60% across multiple locations — Bookafy: smart scheduling with auto-sync that cut no-shows 30% at scale — Cross-document semantic matching with sentence-transformers and FAISS at 93% recall for requirement tracing — Cut irrelevant LLM context injection by 78% with adaptive chunking, and raised generated test-case accuracy to 92% — RAG chatbot for a Swiss banking client and a scalable LLM document-processing pipeline, both 5-star on Upwork 100% Job Success, Top Rated, native English, US business hours. No overnight handoffs, no timezone lag, no waiting until tomorrow for an answer. YOU PROBABLY NEED ME IF — Your RAG pipeline retrieves the wrong chunks and nobody can explain why — A demo works, but there's no path to production, monitoring, or cost control — You need an agent that completes multi-step tasks instead of looping until it burns tokens — Your LLM bill is scaling faster than your usage — You have documents, contracts, or images and need clean structured data out of them — Your n8n, Make or Zapier workflows keep breaking and you want them rebuilt to hold — You've been burned by an AI freelancer who handed over a notebook and disappeared WHAT I BUILD RAG and LLM Search LangChain and LlamaIndex, embedding strategy, chunking, hybrid and re-ranked retrieval, and an evaluation harness so you can prove it got better instead of hoping. Vector stores: Pinecone, FAISS, Qdrant, pgvector. AI Agents and Automation LangGraph, AutoGen and CrewAI when the job needs real engineering. n8n, Make and Zapier when the job is plumbing — with OpenAI and Claude API calls wired in properly instead of duct-taped. Tool selection, multi-step reasoning, guardrails, retries, and hard cost ceilings. Agents that fail loudly instead of silently. Custom ML Models Classification, regression, forecasting, clustering, recommendation, anomaly detection. Scikit-learn, XGBoost, LightGBM, PyTorch, TensorFlow. NLP and Computer Vision Summarization, NER, sentiment, spaCy, NLTK. OCR and document extraction, invoice and form parsing, YOLO, OpenCV. MLOps and Deployment MLflow, Airflow, DVC, Docker, Kubernetes, FastAPI, AWS SageMaker, GCP Vertex AI. CI/CD for models, versioning, monitoring, and a rollback path that exists before you need it. Generative AI Applications GPT-4, Claude, Gemini, LLaMA. Text, image and multimodal, with prompt versioning and evaluation rather than guesswork. INDUSTRIES FinTech and banking, healthcare and MedTech, e-commerce and retail, legal and compliance, real estate and construction, EdTech, and customer support automation. HOW I WORK Weekly demos, not weekly status updates — you see working software, not a percentage. Modular code your team can extend after I'm gone, with documentation and a handover session. Metrics defined before we build, so "done" isn't a matter of opinion. And I'll tell you when a simpler non-AI solution would work better. That conversation has saved clients more money than anything I've built for them. CREDENTIALS MS Computer Science, AI/ML — University of Pittsburgh 100% Job Success, Top Rated on Upwork TELL ME WHAT YOU'RE BUILDING Send me a message with what you're trying to build and where it's stuck. I'll tell you what I'd do first, roughly what it costs, and whether it's worth building at all. If it isn't, I'll say so — that answer is free.