Tayyab M.

AI Engineer specializing in LLM fine-tuning, RAG, and multi-agent systems. Currently shipping production ML at Cointegration, where I built multi-agent workflows with LangChain and AutoGen that cut data retrieval latency by 40%, plus automated reasoning pipelines using Model Context Protocol that replaced 15 hours per week of manual processing. Recent work: Qwen 2.5 7B QLoRA fine-tune on Urdu where v2 hit 66% win rate vs. base across an 8-task evaluation harness. The most useful finding was an 18-point disagreement between Claude, Gemini, and GPT-5 as judges on the same eval, so I treat single-judge evaluation as a smoke test now and use multi-judge medians for final reporting. I work best on: LLM fine-tuning with QLoRA, including eval design and catastrophic forgetting diagnosis RAG pipelines and vector search Multi-agent systems with LangChain, AutoGen, CrewAI, and Model Context Protocol MLOps pipelines with model gating and SHAP explainability Stack: Python, PyTorch, LangChain, AutoGen, CrewAI, OpenAI SDK, MCP, Docker, FastAPI, PostgreSQL, scikit-learn, SHAP.I'm doing my Master's in AI Engineering and available 20+ hours per week. PKT timezone (UTC+5), full overlap with EU clients and morning overlap with US East Coast.