At Lyptus Technologies, I built an agentic AI system that turns natural language specifications into test plans, code, and test suites, then runs and verifies the results.
I designed a Python and FastAPI orchestration engine for specialized LLM agents across OpenAI and Z.ai. I also developed an autonomous repair agent with guardrails that block unsafe fixes.
Through context engineering, I cut prompt size by 96% and peak build memory by 70%; the system is backed by 500+ automated tests.
On Amazon’s AWS Bedrock team, I validated AI/ML datasets, automated data checks with Python on AWS SageMaker, and built AWS QuickSight dashboards. I’ve also developed a CNN for fruit quality detection and a MediaPipe and ROS2 pipeline that maps human boxing poses onto humanoid robots.
