At DeccanAI, I build production AI and backend systems, including a reinforcement-learning platform supporting 500+ annotators and contributing to $10M ARR. I also architected Stark-Forge, a Kubernetes-based distributed backend framework with pluggable verification services.
I delivered 20+ production MCP servers with verified execution traces; the Jira MCP server processes 10,000+ requests daily for evaluation pipelines used by Microsoft and Google. I improved end-to-end task success from 41% to 68% and fine-tuned Qwen-0.5B for text-to-SQL within 2% of GPT-4o-mini accuracy at 10x lower inference cost.
Previously at Deutsche Bank and AiDash, I built ETL pipelines, high-concurrency microservices, and ML tooling that reduced latency, outages, manual GIS work, and experiment setup time.
