fsdl-text-recognizer-2022-labs

Why we included this project

This repository is the lab code for the Full Stack Deep Learning 2022 course. Students build a system that reads handwritten paragraphs straight from images into text, and the work keeps expanding beyond a notebook into a real codebase. It pairs a CNN image encoder with a Transformer decoder in PyTorch Lightning, augments real data with synthetic samples, and includes experiment tracking, testing, and type checking. The value is in watching production concerns get added around the model: TorchScript packaging, a Dockerized predictor on AWS Lambda, a Gradio interface, and monitoring. For developers new to MLOps it offers a rare single end-to-end example, and experienced teams can pull concrete patterns from it.