Sizzle Roasting Software

Built a platform that runs a coffee roastery end to end: bean inventory and costs, live machine control, and AI roast coaching. Designed closed-loop software that automatically steers the roaster to hit a target temperature curve, validated against a physics simulator. Trained a model to predict how a batch will taste from its roast data, then search for the profile with the best predicted flavor. Built a system that hears the beans crack through a microphone and retrains itself on that machine’s own audio. A 47-tool agent on Strands and LangChain runs local models on an NVIDIA Jetson to keep coaching under 8 seconds.