The commonly used linear chat interface doesn't allow you to take quick detours.Copy pasting portion you couldn't understand back into chat interface adds friction. This fixes that, it replaces the linear chat structure with a node based one.
A small LLM inference runtime I built to understand how model serving works under the hood - KV cache, scheduling, batching, streaming, and benchmarking.
It checks your squat, bench, and deadlift form using pose landmarks and a Random Forest model. Gives live feedback & counts your reps. Trained on a custom dataset we collected in our college gym.
