Mechanical engineer who converts CFD/simulation intuition into production AI systems
Mechanical engineer turned AI engineer. I translate CFD/FEA intuition (OpenFOAM, Ansys, Moldex3D, COMSOL) into deterministic AI systems — multi-agent LangGraph + MCP, NVIDIA NIM inference, edge MLOps — and ship them into live production across 9 industry verticals.
Recently live flagship systems
● Live CFD Field · ANSYS Fluent Nucleus
Reactor Coolant Manifold
turbulence: k-ω SST · mesh: 14M cells
Solver Status
Converged
Min residual
1.4e-08
Wall y+
0.92
Iterations
2,148
Click any role to see the full release history and technical stack.
Associate Solution Architect
Architected autonomous multi-agent systems using LangGraph & MCP, edge-first distributed cloud infrastructure, and 3D WebGL digital twins for mission-critical sectors.
The pipeline that fuses mechanical-engineer CFD/FEA intuition with production AI — from domain-meshed geometry all the way to a Llama 3.1 70B agent on Cloudflare Workers. Eight stages, all first-party.
STL/STEP ingestion → auto meshing → hex-dominant refinement on critical surfaces.
Five flagship projects live in production. Each card loads a live preview — click to verify.
Production AI surveillance running COCO-SSD object detection on 16 live DHR Thailand cameras via TensorFlow.js WebGL backend. Live FPS, detection histograms, and auto-alert snapshots.
Interactive previews — click any card to explore the live environment.
Production ML platform monitoring 1,944 UPS batteries across 9 NT data center facilities. Predicts Remaining Useful Life using LSTM neural networks (R²=0.85, MAE=2.3 days) and detects anomalies with Isolation Forest (F1=0.90). Full Medallion data pipeline, TimescaleDB hypertables, LINE Bot alerting with Gemini AI, and 3D facility visualization.
The browser-facing 3D digital twin for NT's 9 data center facilities. Uses Three.js for interactive 3D battery block models, Mapbox for geospatial view of facility locations across Thailand, and real-time WebSocket connections to the Railway backend.
First-generation 3D facility proof-of-concept built to win the NT tender. Uses xeokit SDK with IfcOpenShell for loading IFC 4x3 building models, overlaying architectural BIM geometry with live battery health indicators.
Three-project system for DRMD (กรมฝนหลวงและการบินเกษตร). The mission planner UI shows weather windows, aircraft assignments, and go/no-go confidence scores. A FastAPI backend runs Physics-Informed Neural Networks (PINN) to infer cloud seeding conditions.
Multi-source disaster intelligence dashboard powered by Claude AI. Aggregates data from DDPM (Department of Disaster Prevention and Mitigation) and GDAS feeds, applying LLM-based event classification and priority scoring.
Agricultural field analytics platform that connects to GISTDA satellite geo-data APIs for field mapping. Azure AI handles crop analysis inference. Mapbox renders field polygons and zone analytics, with voice animation for field officers.
Enterprise AI workspace for CPAC Digital Construction built on LangGraph with MCP tool integration and LangFuse V3 observability. POC1 delivers AI Generative Design for retail layout optimization. POC2 maps Revit/IFC to กรมบัญชีกลาง budget codes.
Full-stack embodied carbon intelligence platform calculating lifecycle carbon across stages A1–C4 per EN 15978. A Cloudflare MCP server exposes carbon data as LLM tools, specifically targeting the Thai construction sector.
Enterprise resource planning system for a Thai wood manufacturer: ระบบจัดการต้นทุนไม้ตัวอย่าง. Covers sample request, production planning, inventory tracking, order delivery, and multi-level approvals.
AI-augmented 3D BIM model viewer — type a question, see the affected elements light up inside the model. Built for engineering teams on ASEAN infrastructure projects (hydropower, industrial estates, hospitality).
National-grade weather intelligence platform built entirely on Cloudflare's edge. Runs FourCastNet and GraphCast ML inference directly in Workers AI, with an OpenTelemetry distributed tracing mesh across 9 dedicated Workers.
Full-stack restaurant intelligence platform for the F&B sector. Covers location intelligence, competitor analysis, and menu optimization through a LangGraph multi-agent backend spanning 2 Cloudflare accounts.
A Cloudflare Worker protocol translator between NVIDIA NIM API format and Anthropic Server-Sent Events (SSE) streaming format. Confirmed working at 29ms TTFT, enabling Claude to be used inside enterprise VSCode toolchains with zero extra infra cost.
Enterprise AI document editor where three specialized agents work concurrently via MCP orchestration: a Reviewer flagging issues, a Writer rewriting sections, and an Analyst extracting insights. A Cursor-like experience for Word documents.
AI-augmented property management system for AOT's real estate and facility portfolio. Using Noto Sans Thai for native bilingual interfaces, connected to FastAPI and Cloudflare Worker edge APIs for anomaly alerts and property health scores.
AI-augmented SCADA interface styled after AVEVA design tokens (#0097D1). Displays real-time P&ID diagrams, equipment status, sensor trends, and AI-generated anomaly alerts for NT facility operations teams monitoring power and cooling.
PolyGuard SCADA is a next-gen SCADA platform for plastic injection molding in the food packaging industry — targeted at the cleaning/diagnostic workflow that C.P. Food Packaging Industry pioneered. Real-time process telemetry with machine learning-based cleanup-quality scoring and production forecasts.
Civic technology project for blood donation coordination in Thailand. A LINE Official Account chatbot handles donor registration, blood type matching queries, and urgent shortage alerts, with gamified civic engagement (BloodPlus Fight).
FacilityHub is a building management platform with an embedded AI surface at /ai — surfaces CCTV feeds, access logs, floor plans, and predictive maintenance insights in one operator-grade interface. Built by a mechanical engineer with CFD expertise who applied his simulation domain knowledge directly to facility operations.
The Earth-2 Digital Twin is a real-time climate simulation cockpit where GraphCast, FourCastNet, DLWP and ENS ensembles drive a 3D Earth. Built by a mechanical engineer with domain CFD background who translated the simulation ↔ data ↔ visualisation axiom from OpenFOAM / Ansys / Moldex3D work into a production geospatial AI surface.
Ensimu.studio is my flagship CFD + AI work. 10 LangGraph agents orchestrate mesh generation, boundary conditions, solver tuning (OpenFOAM), and post-processing (Trame professional workbench — COMSOL-grade 3D viz). I'm using my mechanical-engineer domain knowledge — CFD, FEA, Moldex3D, COMSOL — to teach the agents physical causality, not just pattern recognition.
A live graph of every technology Khiw deploys in production.
Newest AI Engineer tooling breakdowns, autonomous multi-agent workflows, and SEO updates.
★ FEATURED AI GUIDEExplore how Forward Deployed AI Engineers solve complex enterprise bottlenecks by combining LangGraph supervisor architectures with Model Context Protocol (MCP) servers.
Edge AI & Cloud MLOpsDiscover how to engineer an ultra-low latency Cloudflare Worker protocol proxy that translates NVIDIA NIM API calls to Anthropic SSE format in 29ms.
3D Spatial & Digital TwinsLearn how to transform millions of raw IoT sensor telemetry events into interactive, real-time 3D spatial digital twins for industrial refinery and telecom facilities.
Data Engineering & MLOpsDeep dive into organizing landing, bronze, silver, and gold time-series data pipelines inside TimescaleDB and Azure Synapse for high-accuracy RUL prediction.
Every shipped engineering milestone, versioned and dated across all roles.
Full-stack range with an AI-native core — from LangGraph/MCP to 3D Digital Twins.
Have a complex enterprise challenge needing forward-deployed AI, data pipelines, or 3D spatial intelligence? Let's connect.