Lately I've been knee deep on some machine learning problems while building NeuroType, my typing practice and memory training app.
The big one: tuning a spaced repetition model that actually adapts to a user's real recall curve instead of a fixed interval schedule. Getting the decay parameters right from sparse early data is harder than it looks.
I also hit classic issues around feature selection, deciding which typing signals (latency, error clusters, correction patterns) actually predict retention versus just adding noise, and keeping the model lightweight enough to run client-side without a training pipeline behind it.
More notes soon as I iterate. If you've wrestled with adaptive scheduling or on-device inference tradeoffs, I'd love to compare approaches.AI systems thinker focused on automation, structured data, and scalable workflows.
Top comments (0)