Anhar Munir.

Developed a complete AI-powered smart glasses system for visually impaired individuals integrating YOLOv8-based indoor and outdoor object detection, achieving real-time inference on edge hardware.

• Built Pakistani currency recognition module with 99.9% accuracy and Tesseract OCR for real-time text reading from documents and signage.
• Implemented offline GPS navigation, live location sharing, and a fully voice-controlled interface using Vosk STT and pyttsx3 TTS.
• Optimized and deployed all AI modules on Raspberry Pi 4B using TFLite INT8 quantized models, achieving real-time edge inference within power constraints.

Developing an AI-powered road safety system that detects vehicles in foggy conditions via a mobile app integrated with a forward-facing camera and deep learning models hosted on Hugging Face.

• Designed an image dehazing pre-processing pipeline using OpenCV to enhance visibility before feeding frames into the YOLOv8 detection model, improving inference accuracy in adverse weather.
• Integrating an ultrasonic distance sensor to estimate proximity of detected vehicles and trigger real-time audio alerts through the car's speakers for driver safety.
• Targeting practical affordability and retrofittability — the system requires no vehicle hardware modifications beyond a mounted camera and smartphone.

Built a full-stack AI web application for breast cancer detection using a fine-tuned Swin Transformer trained on the CBIS-DDSM mammography dataset, achieving 87.5% accuracy and 0.884 ROC-AUC.

• Implemented a dual-view ensemble analysis system (CC + MLO views) mimicking clinical radiologist workflow, reducing false positives through probability averaging.
• Developed a FastAPI backend with optimized model inference and caching, paired with a React frontend featuring drag-and-drop image upload, real-time results, and automated PDF report generation.
• Applied transfer learning, class weighting, and data augmentation techniques to address medical imaging challenges on ~30,000 image dataset.