• 🎓 Undergraduate in B.Sc (Hons) Electronics & Computer Science at University of Kelaniya
• 🤖 Passionate about AI/ML, Computer Vision, Robotics, IoT, and Embedded Systems
• 🔧 I bridge hardware ↔ software: microcontrollers, sensors, edge inference, and real-time dashboards
• 🏊‍♂️ Outside tech: National Swimmer

• 🔎 Detection & Classification — YOLOv8, Faster R-CNN, OpenCV
• 🧾 Text & OCR — Tesseract, EasyOCR, custom post-processing
• 🎥 Image/Video — augmentation, illumination handling, defect analysis
• ⚡ Optimization — quantization, pruning, ONNX/TensorRT for real-time
• 🛠 Hardware Integration — ESP32/edge, Jetson/RPi, PLC comms
• 🖐️ Human pose/hand — MediaPipe pipelines & calibration UIs

• 🫖 Tea Packaging Defect Inspector
  Real-time defect detection for tea packets (leaks, seal gaps, misprints). PLC-ready reject logic, OCR for date/lot codes.
  Tech: Python YOLOv8 OpenCV Tesseract PLC ESP32

• 🪪 License Plate Blurring + OCR Pipeline
  Privacy-first video pipeline: YOLO detection, tracking/debounce, OCR, clarity scoring, selective blur, analytics, CSV/JSON exports.
  Tech: Ultralytics YOLO OpenCV supervision FastAPI (planned)

• ✋ Finger Ability Tracking & Rehab Metrics
  MediaPipe-based 21-landmark tracking with calibration to 0–100% (neutral ↔ max), clamped UI for out-of-range values, per-digit graphs.
  Tech: MediaPipe OpenCV NumPy Python

• 🏠 Digital Twin Smart Home (Flutter)
  Device tiles (switches, sensors), alerts feed (e.g., “Bedroom Light ON”), project-centric overview (see Projects Hub below), and room cards with temps/setpoints.
  Tech: Flutter Dart

• 🎭 Facial Landmark 3D Projection
  Real-time 2D→3D mapping + Blender/Matplotlib visualization.
  Tech: Mediapipe OpenCV Blender

• AI-Driven Tea Package Defect Detection – Concept Notes

  • Defect taxonomy: seal gap, tear, underfill, misprint, date/lot OCR fail
  • Pipeline: capture → pre-proc → detection (YOLOv8) → OCR → PLC signal (reject)
  • Metrics: FP/FN, read rate, throughput @ latency budget, MTBF
  • Edge concerns: lighting normalization, motion blur, exposure control
  • Future: few-shot adaptation per SKU, synthetic data for rare defects

• Tooling

  • Labeling: Roboflow/Label Studio
  • Training: PyTorch/Ultralytics, augmentation recipes
  • Export: ONNX/TensorRT (where applicable)
  • HIL tests: looped conveyor footage + threshold sweeps

Add an assets/ml_pipeline.png diagram/GIF showing collect → label → train → evaluate → deploy.

• 🧪 Refining Tea Defect Inspector models + PLC interfacing
• 📡 Building LoRa/ESP32 sensor nodes and analytics
• 🧰 Improving Flutter dashboards and alert UX

• 💼 Treasurer – IEEE IES Student Chapter, University of Kelaniya
• 📌 Project Manager – UOK Robot Battles 2k25

• 💼 LinkedIn: supun-tharaka-6bb8b5278
• 📧 Email: [email protected]