Engineer who has shipped in two separate domains: manufacturing digital twins on NVIDIA Isaac Sim (1 yr 1 mo) and real-time industrial-safety computer vision with TensorRT (1 yr 9 mo). 2 yrs 10 mo total.

• Digital twin: text-free drawing (DXF) semantic-recovery pipeline; auto-matching search space 424,270 → 70 (~6,000×); upload 62 s → 4.8 s, save 153 s → 14 s.
• Computer vision: TensorRT + dynamic tensor memory took Human Pose Estimation from 17–19 FPS to 40–50 FPS.
• Document AI (Aug–Oct 2025, before the digital-twin work): based on an in-house prototype, extended a body-part dimensional-quality (T/O·DIM'S) web app — upgraded its multimodal-LLM inspection-report readout to Gemini 2.5 Pro and added a predicted-dimension 3D convex-hull viewer; built and tested an in-house RAG chatbot over standard-work manuals (LlamaParse → Pinecone, LangChain + Gemini) that cites source document and page — ~10 test documents, LangGraph migration under review when the project was closed.
• Work code lives in private company repositories and is not reflected in this account's contribution graph.

I build AI Agents for industrial workflows — bringing LLM-based agent systems into the on-the-ground constraints of automotive manufacturing, robotics, and digital-twin simulation, and making shop-floor documents (manuals, inspection reports, drawings) machine-readable with LLMs and vision.

• 🔭 Currently studying: agent harnesses, MCP, evals
• 🧰 Most recent role: Industrial digital-twin solutions delivered into the automotive Tier-1 supply chain
• 🧪 First projects there (Aug–Oct 2025, before the auto-layout digital-twin system of Nov 2025 – Apr 2026): two in-house prototypes — inspection-report LLM readout for a dimensional-quality web app (based on an in-house prototype), and a standard-work-manual RAG chatbot
• 📫 Reach me: qowotjd2010@gmail.com

_{Currently studying agent harnesses, MCP, and evals.}