As an independent AI Trainer, Prompt Engineer & Evaluation Specialist, I curate prompt-response datasets and preference rankings using RLHF/DPO concepts. I also create multi-turn reasoning traces and edge-case examples to help reduce hallucinations.
I authored coding and reasoning benchmarks based on open-source repositories such as sqlglot, including hidden test suites and reference solutions. For evaluation, I deployed reproducible harnesses with Docker and VMware Fusion on macOS.
I developed an autonomous quantitative trading agent that uses Gemini LLMs for market sentiment analysis and executes mathematical indicators. Its tool-calling pipelines use schema constraints and volatility-adjusted risk management.
I also architected a generative audio and story web application for Hindi narratives, and open-sourced OrbitKit, a 12-component UI library with GitHub Actions CI/CD and 100% test coverage.
