On the line I commission, tune and troubleshoot; in the code I build workflow automation and AI.

A full HMI product with a built-in AI assistant: screen designer, machine runtime and PLC drivers

A PC-based alternative to dedicated HMI panels, built as a product to sell. Engineers draw machine screens in a Windows designer, or describe them to the built-in AI assistant in English or Malay, which plans the screen, binds it to the project's real PLC tags and builds it for the engineer to check. The screens compile into a signed bundle and deploy to a Windows machine PC, where a runtime service talks to the PLCs, logs alarms and history, and reports to a central server. One PC can serve several machines.

A Telegram bot that runs a support team's daily operations

A single Telegram bot that replaces the spreadsheets, sticky notes and group-chat searches a field-support team runs on. From one chat, engineers log their daily work, raise and track tickets across the team, look up spare parts, and open documentation. Everything is backed by Google Sheets and Google Drive, and entries are auto-translated to English.

An industrial machine-vision system that tunes itself

Proof of concept, still under active development. Features and results may change.

An industrial machine-vision system that tunes itself: a toolbox of selectable vision Tools (OCR, defect detection, dimensional measurement, template matching) running under a single loop that improves each Tool from a verification signal instead of relying on hours of manual tuning. Inspired by autonomous-research loops, applied to factory inspection.

A 24/7 trading-signal system that reasons with Claude AI

An autonomous trading-signal bot that scans US equities, extracts technical features, and asks Claude AI to classify setups into BUY / WATCH / AVOID. Signals pass through deterministic risk gates, get paper-traded automatically, and their outcomes feed a self-learning engine that recalibrates win rates daily. It runs around the clock on a mini PC and is operated entirely from Telegram.

One Windows app that unifies a robot and a laser profile sensor

JR3303 Studio: a native Windows app (C# / .NET, WPF) that unifies a Janome JR3303 desktop robot and a wenglor weCat3D laser profile sensor into one inspection cell. Instead of juggling two vendor tools, the operator positions a part with the robot, reads the live laser cross-section, measures it on screen, and every result is logged to a CSV. The robot's Ethernet protocol was reverse-engineered from scratch; the sensor streams 1280 points per profile, calibrated to millimetres.