AI Platforms
AI assistants for the lab and the factory: RAG over technical documents, Document Intelligence, AI Search, long-term memory for agents.
Principal Engineer at Milwaukee Tool (TTI Group). Architecting AI platforms, IoT pipelines from sensor to cloud, and industrial test automation.
I started as a technician on a Tetra Pak packaging line. More than 7 years later I am responsible for the technical direction and architecture of 36+ production systems. Six areas, one engineer.
The whole chain I design and run. Data goes up from instruments and edge devices. Decisions come down from the platform and the AI.
AI assistants for the lab and the factory: RAG over technical documents, Document Intelligence, AI Search, long-term memory for agents.
Firmware for industrial edge gateways, field protocols, and OTA updates for fleets of hundreds of nodes.
A background from the production line: Siemens and Mitsubishi PLCs, SCADA, HMI. I learned the machines before writing software for them.
Test rigs and life tests, high-speed data acquisition, a Lab Management System for product testing labs.
Multi-module web platforms with enterprise SSO and real-time WebSocket/SSE, packaged and run in containers.
Time-series, graph and analytics: InfluxDB on the lab floor, then a data lake, Databricks and Kafka for the whole plant.
Each project page is built by AI from the GitHub or Azure DevOps repo.
A digital transformation and AI platform for one test lab in Vietnam. The lab is small but runs like a complete organization: test engineers, the test automation group, purchasing, planning, warehouse and operations.
Manages the full supplier lifecycle, with real-time updates over WebSocket/SSE and Azure AD SSO login.
Firmware for industrial edge gateways running FreeRTOS, with I/O configurable per machine type.
Industrial test application that acquires machine data, stores it in InfluxDB, shows it in Grafana and archives to Azure Blob.
A Grafana plugin that shows a building floor plan with real-time sensor data.
Python tools for engineers, including the Striker Speed Profile Analyzer, which uses computer vision and optical flow.
AI agents turn notes, commits and operating data into posts. A reviewer agent checks the numbers and code before a post goes live.
A NeurIPS 2024 benchmark re-scored 40 anomaly detectors on 1,070 curated time series, and a subsequence PCA came out on top of the univariate table.
In dkhoa-dev/nestjs-multer-gridfs, downloads stream but uploads buffer the whole file. Here is where, and what to change.
Happy to talk about AI platforms, industrial IoT, test automation and plant infrastructure.