What I do
The problems I keep coming back to, in the lab and in the field.
I work across the full vision stack. That starts with data — collecting from deployment cameras, auditing labels, and training detectors that hold up on footage the benchmarks never saw.
Then comes inference where it actually runs: ONNX pipelines, careful quantization, and detect-sparse/track-dense designs on Jetson and CPU edge boxes. And around all of it, leading teams that take AI systems from architecture to production — CCTV intelligence, industrial sortation, and applied research that ships.
- Training — field datasets, honest validation, models that survive deployment.
- Inference — real-time perception on the hardware at hand.
- Shipping — teams, pipelines, and systems in production.
Research
Physics-rooted questions I keep investigating.
Before production systems, there was a photonics lab: optics, signal processing, and computational physics applied to real measurement problems — including laser cross-section volumetry for my undergraduate thesis at Universitas Riau. That foundation still drives how I think about imaging: physics first, models second.
Lately that means spectral analysis — latent-variable methods like PCA and PLS for pulling chemistry out of high-dimensional spectra — alongside field machine vision that has to survive orchards, ports, and factories, where the lighting never cooperates. I wrote up the spectra side as a primer: Latent space for chemometrics →
Writing
Notes on vision systems, inference, and training.
All posts →Get in touch
Industrial sortation, CCTV intelligence, edge perception, or research collaboration.