About this site

rlcapstone.ai is a collection of capstone-style AI projects, each built completely: dataset preparation, model training on local GPU hardware, evaluation with honest metrics, and deployment into a real working application.

The write-ups aim to be the article I wished existed when starting each project. That means including the parts tutorials usually skip — export toolchains that only work on certain platforms, dependency versions that silently break builds, evaluation thresholds chosen for a reason, and results that are good but not magic.

How projects are built

• Training — PyTorch on a local NVIDIA GPU, using public research datasets.
• Deployment — on-device inference wherever possible (TensorFlow Lite, ONNX Runtime), so demos work offline and no user data leaves the device.
• Apps — Flutter for mobile, plain web for demos, one codebase per project.
• Honesty — every project page reports real metrics on held-out data and states its limitations plainly.

Contact

Questions or corrections are welcome — reach out via the project repositories on GitHub.