Vedant Asawa

I’m an AI/data science evaluation engineer focused on making LLM and ML workflows measurable, reliable, and repeatable. At Xlairs, I created 50+ evaluation problems across statistical analysis and forecasting, validated 100+ LLM solutions with rubric-driven checks, and identified failure patterns that improved evaluation accuracy by 30%.

In a Turing LLM Engineer (Contract) role, I authored 500+ multi-turn datasets for tool-augmented LLM supervised fine-tuning (SFT), implemented JSON function-calling workflows for API tool chaining, and applied RLHF to reduce hallucinations and improve tool selection—achieving a 98%+ approval rate. I also build end-to-end systems, from a MERN file-sharing application with a 99.9% upload success rate to a TensorFlow satellite pollution model reaching 88% accuracy, supported by robust validation for generalization.