Software engineer working across AI systems, robotics, and product.

Building a serverless agent system that classifies, diagnoses, and drafts resolutions for ServiceNow tickets from a 150+ user data platform processing 8,000+ jobs each day.

• Routes six root-cause categories through EventBridge and Lambda to a Bedrock supervisor that delegates retrieval and diagnosis work.
• Uses Step Functions and API Gateway callbacks for human approval before writing resolution notes to ServiceNow or updating the knowledge base.
• Reduced diagnosis from roughly 60 minutes to under 5; designed the routine-ticket path to resolve about 40% of request work.
• Added Dynatrace/Graylog observability and dev/UAT/prod delivery through EKS, Terraform, and GitHub Actions.

Building a requirements engine that turns raw notes and transcripts into technical specs, linked product artifacts, and interactive prototypes.

• Maintains an artifact graph of 50+ cross-linked PRD sections, user stories, test cases, timelines, and prototype blocks.
• Uses schema-constrained outputs and a PatternRegistry so generated designs stay inside valid application patterns instead of drifting between edits.
• Applies atomic patches that map visual blocks back to PRD section IDs, updating only the affected requirements and prototype components.
• Serving 3 pilot teams; backed by $25K in Microsoft for Startups funding and named a Klaus Startup Challenge finalist.

Leading the 15-person team turning Graption from a hackathon prototype into a persistent accessibility platform for speaker-aware captions and missed-context recovery.

• Re-architected the product around React Native, FastAPI, WebSockets, and PostgreSQL for real-time mobile caption delivery.
• Combines OpenCV/MediaPipe visual features, Whisper transcripts, and audio events to identify who spoke and where they were located.
• Reached 0.74 F1 on 8 hours of audio-video at 30 FPS and under 1.5 seconds latency; conversation retrieval reconstructs context a user missed.

Automated recurring revenue and billing workflows in Oracle ORMB through Python ETL pipelines on AWS Lambda and S3.

• Built reconciliation checks and staging-data validation that caught incomplete or inconsistent records before they reached production billing.
• Trained a random forest anomaly detector with 0.88 precision and 0.79 recall, surfacing 21 misbilled records.
• Implemented Spring Boot/MySQL pricing APIs plus SQL and ORMB configuration templates for new billing scenarios.

Built a ROS2 and MuJoCo spacecraft simulator coordinating 8 agents at 500 Hz with physics, GNC, onboard vision, and formation control in the loop.

• Generated structured simulation rollouts for decentralized learning rather than relying on expensive physical test data.
• Trained coordination models across roughly 500 rollouts, reducing simulated fuel cost by approximately 20% while maintaining collision-free trajectories.