Amirah Heng

Developed advanced computer vision solutions, including face, ID, and fingerprint liveness detection models using a YOLO+DINOv3 hybrid approach with 99.8% accuracy across paper and screen replay attacks. Implemented aggressive data augmentation and multi-phase training to optimize performance across challenging imaging conditions. Built biometric fraud checks in the backend, exposed REST APIs, and integrated Python-based models via ONNX Runtime and inter-process communication. Translated Python CV image processing to optimized C++ and integrated ONNX models into Android applications for on-device, real-time detection of ID, passport, and face spoofing. Introduced SAM3 labeling methodology from Meta to speed up data labeling and produced automated batch testing, confidence analysis, and visualization workflows. Productized a mobile OCR application by translating Dart CV logic into optimized C++ in collaboration with perception engineering teams.

Led Agentic AI chatbot POCs for manufacturing using Google’s ADK and Vanna AI as a database query agent. Conducted end-to-end technical sales presentations and product training for AI-based visual inspection solutions, engaging C-level and manufacturing operations leaders.

Built a 30-day price-forecasting ML pipeline for 100+ SKUs achieving 90%+ accuracy using AutoML, MLFlow, Feature Store, SKForecast, Redis, and AWS Redshift/S3. Optimized LLM performance through prompt engineering and fine-tuning for agricultural use cases such as pest control management.