I build applied AI systems, research tools, and developer products.
Interests: time-series foundation models; LoRA and fine-tuning; forecasting and calibration; reproducible model evaluation; dataset discovery, preprocessing, and provenance; AI agents and developer tools; synthetic populations and simulation.
Public projects include Chronos-2 DE-LU electricity forecasting, Automated Finding v2 for IRW dataset ingestion and evaluation, Promptr for VS Code and Cursor, and AudienceTwin, my fork of sim francisco.
I care about strong baselines, time-ordered evaluation, useful release artifacts, and clear reporting of what worked and what did not.
Latest model: https://huggingface.co/AryanSudhirDev/chronos-2-de-lu-pregate
Code and research: https://github.com/AryanSudhirDev