What I do · Demos · Projects · How I build · Stack · Connect
I'm an AI Engineer in Computer Vision, Deep Learning, OCR, ANPR and Edge AI. I build vision systems for real industrial use, the kind that have to work on blurry, angled, badly lit camera feeds, not just clean datasets.
Data → Annotation → Preprocessing → Training → Evaluation → Optimization → Deployment
🟢 Open to roles in AI Engineering, Computer Vision, Deep Learning, Machine Learning and Edge AI.
|
YOLOv8 and YOLO11 models for objects, people, vehicles, PPE, fire and smoke. Tracking across frames and zone-based alerts. |
ANPR and OCR with PaddleOCR. Deskewing, enhancement, multi-frame voting and Indian plate regex validation. |
ONNX Runtime optimization, NVIDIA Jetson and RTSP / CCTV streams, with events and alerts stored in SQLite. |
ROI monitor · ANPR pipeline · Metrics lab
Click the image to open the full interactive portfolio.
🚘 ANPR / License Plate Recognition · YOLO11 PaddleOCR OpenCV
Vehicle → plate → OCR, built for blurry, angled, real CCTV frames.
flowchart LR
A[CCTV Frame] --> B[Vehicle]
B --> C[Plate]
C --> D[ROI]
D --> E[Deskew + Refine]
E --> F[Enhance]
F --> G[OCR]
G --> H[Multi-frame Vote]
H --> I[Regex Validate]
I --> J((Final Plate))
- Perspective correction and deskewing
- Multi-frame OCR voting
- Indian plate regex correction
- Blur, contrast and brightness checks
📦 Industrial Object Detection · YOLOv8 Roboflow PyTorch
- Sack bag, tin and carton box detection
- Plywood detection
- Dataset engineering in Roboflow and accuracy tuning for real plant lighting
👥 Person Detection & Counting · YOLO11 Tracking OpenCV
- Person detection with object tracking across frames
- Counting with zone lines
- Real-time video analytics
🦺 PPE Detection & Compliance · YOLOv8 Roboflow PyTorch
- Helmet and safety equipment detection
- Dataset balancing, annotation review, hard-negative analysis
- Compliance monitoring and alerts, retraining after error review
🔥 Fire & Smoke Detection · YOLOv8 RTSP Jetson
- Real-time detection with instant alerts
- False-positive analysis and retraining
- Deployment testing on live streams
🚗 Vehicle Detection & Classification · YOLO11 ONNX Runtime
- Multiple vehicle classes on CCTV streams
- Camera angle and positioning analysis
- Model performance evaluation
📍 ROI / Zone Monitoring & Alerts · OpenCV SQLite JSON/YAML
- Per-camera ROI and zone rules
- Configurable pipelines in JSON / YAML
- Events and alerts stored in SQLite
flowchart TD
A[Dataset collection + cleaning] --> B[Annotation + validation]
B --> C[YOLO training]
C --> D{Evaluate: P / R / F1 / mAP}
D -->|errors found| E[Error analysis: false positives, hard negatives]
E --> C
D -->|good enough| F[Optimize: ONNX, memory, latency]
F --> G[Real-time testing on live streams]
G --> H[Edge deployment]
I care about what happens after training. Camera position, image quality, preprocessing, dataset quality and inference speed often matter as much as the model itself.
Skills by area
| Area | What I do |
|---|---|
| Detection | YOLOv8 / YOLO11, object tracking, real-time video analytics |
| OCR / ANPR | PaddleOCR, OCR post-processing, plate recognition, multi-frame voting |
| Data | Roboflow, dataset engineering, annotation review, class balancing, hard-negative analysis |
| Optimization | ONNX Runtime, model optimization, memory and latency tuning |
| Edge | NVIDIA Jetson, RTSP / CCTV streams |
| Monitoring | ROI and zone-based monitoring, alert systems |
| Pipelines | Configurable AI pipelines in JSON / YAML, SQLite event storage |
Have a vision problem that has to work on real cameras? Let's talk.