Available for freelance & full-time work · Colombo, Sri Lanka
Hi, I'm
AI/ML Engineer
I build AI systems that run on real hardware and in real factories, from a voice-driven smart cane for the visually impaired to a live production monitoring system for a garment manufacturer.
Who I am
I'm a final-year(4th) BSc (Hons) Electronics & Information Technology student at the University of Colombo. I work where machine learning meets embedded systems and industrial automation, building systems that run on real hardware rather than only in notebooks.
My projects include a smart cane with an edge AI sign-recognition model on an ESP32-CAM, a private RAG document assistant that keeps confidential questions on a local model, an end-to-end MLOps system for machine failure prediction, and a production line monitoring system I built and deployed for PBK Apparels.
I lead the IT team at the Colombo Electronics & Robotics Club, where I run the technical side of our workshops and competitions. I'm bilingual in English and Sinhala.
What I work with
Also experienced with
Selected work
A web app I built solo and deployed for PBK Apparels that runs their sewing lines from quote to packing, with costing, an interactive line map, printable floor documents and a live Director dashboard. It is in live use and replaces daily Excel workbooks.
A team project I led: a smart cane and mobile app that helps visually impaired people navigate indoors and outdoors by voice. I built the AI assistant and the quantized sign-recognition model on the ESP32-CAM, and integrated them into the cane.
A privacy-first RAG assistant that answers questions from company documents with source citations. Admins choose a cloud or local model, and questions touching confidential documents are answered only by the local model, with no cloud fallback.
An MLOps system that predicts industrial machine failures, covering a reproducible pipeline, experiment tracking, a model registry, ONNX serving and drift monitoring. It reached 0.835 PR-AUC and caught 60 of 68 failures on a held-out test set.
A licence plate recognition pipeline for Sri Lankan vehicles, using a fine-tuned YOLOv8 detector, OpenCV preprocessing, OCR and a format validator. A video mode tracks plates across frames to filter out single-frame errors.
An Android app where drivers find and reserve parking on a live map, and lot owners verify and manage spots with QR check-in. Only spots with a verified property deed are shown to drivers.
Career & education
2026Current
Colombo Electronics & Robotics Club (CERC), University of Colombo
I lead the IT team and manage the club's technology infrastructure. I run the technical side of its robotics and electronics workshops and competitions.
Current
PBK Apparels
I designed, built and deployed a production line monitoring system on my own. It replaced daily hand-copied Excel workbooks with one live system used by the Industrial Engineer, the Recorder and the Director.
Surge Robotics (Idea8), Sri Lanka
I annotated human body keypoints and sleeve, trouser and body landmarks for garment-measurement datasets used to train production computer vision models. I kept labelling consistent across large datasets.
Expected May 2027Current
University of Colombo
Coursework in Machine Learning, Deep Learning, Computer Vision, Robotics & Automation, Industrial Automation (A grade) and Advanced Database Systems.
What people say
It finally gave us one clear picture of our production floor.
For years, PBK Apparels ran its production lines on Excel workbooks. The operation breakdowns, dividing plans, hourly output sheets and costing sheets were all separate files kept by different people. By the time a problem on the line reached my desk, the day was usually over, and we were explaining yesterday’s shortfall instead of fixing today’s.
This system changed that. Our industrial engineer now imports the buyer’s OB sheet directly, plans the manpower from pitch time, and sets up the line on a map that looks exactly like our floor: every machine, every operator, and who is doing which operation. When a line is out of balance, the system suggests a plan for the people we actually have that day, and it does so the way our supervisors would: the same machine, the next job in the same section, nobody walking across the line. Our IEs trust the suggestions because they make sense on the floor.
Our recorders enter each operator’s output every hour, with a reason whenever someone is below target, and the day’s cutting at the end of the shift. From the dashboard I can see every line’s efficiency, the bottleneck, the reasons for lost hours, and exactly how far each order has to go, from cutting through to packing. I no longer wait for the evening report. I can see it as it happens, even from my phone.
The printed sheets matter just as much to us. The operation breakdown, the line map for the supervisor, the workstation cards at each machine and the costing sheet we send to buyers all come straight from the system, consistent and ready to sign. Our costing is now based on real pitch times, and our quotes go out faster and with more confidence.
Since we moved to the system, hours lost without a reason have become rare, changeovers are planned rather than improvised, and our meetings start from the same numbers instead of arguing about whose spreadsheet is right.
What I value most is that it was built around how a Sri Lankan garment factory really works, by people who listened. Every piece of feedback from our IEs, recorders and managers was turned into an improvement, quickly. It has become part of how we run PBK Apparels every day.
Harsha Nissanka
CEO,
PBK Apparels
Get in touch
Have a project, a role or a question? Send me a message and I'll get back to you.