Project
Fincheck
Fincheck is a risk-aware handwritten digit validation framework designed for financial systems such as cheque processing. Rather than always producing a prediction, the system estimates confidence and uncertainty, computes financial risk metrics, and rejects ambiguous inputs to reduce false acceptance and false rejection risks. The platform combines deep learning, computer vision, evolutionary optimization, and interactive visualization into an end-to-end evaluation pipeline.
Technologies
PythonFastAPIPyTorchOpenCVNext.jsTypeScriptTailwind CSSMongoDBBun
Key Features
- Confidence-aware handwritten digit validation
- Multi-model CNN evaluation and comparison
- Financial risk analysis using FAR and FRR metrics
- Evolutionary optimization for adaptive risk calibration
- Interactive dashboard with PDF reporting and experiment logging
Engineering Highlights
- →Designed a confidence-aware validation pipeline instead of a traditional OCR system.
- →Implemented Evolutionary Risk Optimization to learn optimal FAR/FRR weighting.
- →Benchmarked compressed CNN architectures on MNIST and CIFAR for robustness evaluation.