Sushant Kumar

Engineered multi-step agentic workflow systems integrating LLM-driven decisioning and API execution, improving task completion rate by 35% and reducing manual merchant operations across platform workflows. Built end-to-end orchestration pipelines with structured prompt-to-API execution, reducing workflow latency by 25% and increasing automation reliability through optimized state management and retry handling. Implemented evaluation frameworks combining LLM-as-judge and human feedback loops, accelerating model iteration cycles by 40% while improving response accuracy and reducing hallucination rates in production systems.

Engineered end-to-end AI-driven data platform integrating real-time pipelines and ML inference APIs, improving prediction accuracy by 35% and reducing decision latency by 40% across enterprise workflows. Developed scalable microservices architecture for ML model deployment, increasing system throughput by 50% while reducing API response time to sub-200ms through optimized request handling and caching strategies. Designed and implemented RESTful APIs using Python and FastAPI, ensuring high availability, fault tolerance, and idempotent service design aligned with distributed microservices architecture principles.