About Me
I'm a researcher and AI systems developer studying Computer Science & Mathematics at Stanford. I like working on ideas that push boundaries and then turning them into real systems people can actually use. My work covers computer vision, multi-agent planning, optimization, and edge AI, and I draw on both theory and hands-on software engineering to make it happen.
I've built everything from high-performance model architectures to real-time AI pipelines, always aiming for efficiency, scalability, and reliability. I enjoy connecting different areas of research and engineering to solve tough problems in practical ways.
Right now, I'm at Salesforce AI, working on agent-graph optimization. My focus is on finding faster ways to run multi-agent workflows by combining structured optimization techniques with LLM-driven strategies. Whatever the project, I aim to build AI that is not only impactful, but also accurate, deployable, and ready for real-world use.
Technical Skills
Languages:
ML / Data:
Optimization / Modeling:
Systems / Tools:
Hardware / Design:
Education
Stanford University
B.S. in Computer Science, Artificial Intelligence Track• Jun 2027 (expected)
Experience
My research and professional experience across AI systems, computer vision, and environmental monitoring
AI Software Engineer Intern
June 2025 – Sep. 2025Salesforce AI Research
•Palo Alto, CAPart of the Atlas AI research team, building agent-graph optimization frameworks.
Technologies
Key Achievements
- Built a modular agent-graph optimization framework for automating multi-agent planning workflows
- Researched symbolic graph optimization methods including evolutionary algorithms and LLM-guided descent
- Designed and implemented graph rewrite tools for subgraph collapsing, node pruning, and operator selection
- Analyzed optimization performance across task types via ablation studies and generalization testing
Student Researcher
Jan. 2025 – PresentStanford Intelligent Systems Laboratory
•Stanford, CAWorked on ground station placement optimization for satellite communication networks.
Technologies
Key Achievements
- Built MILP-based models in Pyomo to optimize satellite ground station placement across mission objectives
- Used TLE data with the Brahe astrodynamics library to generate satellite–station visibility constraints
- Developed Python analysis tools to evaluate data throughput and identify communication coverage gaps
Co-founder and CTO
Oct 2021 – PresentIyarkai LLC
•Washington StateFounded Iyarkai to develop AI tools for wildlife monitoring and conservation; secured a contract with WSDOT.
Technologies
Key Achievements
- Created a morphing pipeline to simulate thermal data from 26K+ labeled COCO-format optical images
- Built a two-stage YOLOv8 pipeline combining synthetic trained detection and real thermal classification
- Achieved >97% precision and recall on real-world deployments across I-90 camera traps
- Deployed system across active WSDOT sites; expansion planned for 12 additional highway crossings
- Presented methods and system design at ICOET & TWS Western Section Conference 2025
Applied Science Intern
June 2021 – June 2025UC Davis Road Ecology Center
•Davis, CADeveloped RADIS: a YOLO-based edge-computing system for real-time wildlife detection and classification to detect animals on highways and warn drivers of active animal presence.
Technologies
Key Achievements
- Trained models on custom-labeled roadside datasets to detect animals in varying highway environments
- Deployed RADIS on NVIDIA Jetson devices in Nevada with NDOT; achieved 97% detection accuracy and 99% system uptime
- Benchmarked inference latency and power consumption to meet real-time roadside alerting requirements
- Implemented algorithms to track animal speed and direction to identify potential collision threats
- Presented findings at ICOET 2023
Engineering Intern
Jan. 2022 – Aug. 2022UC Davis Veterinary School of Medicine
•Davis, CAWorked on the California Mountain Lion project developing animal deterrent systems for wildlife conservation.
Technologies
Key Achievements
- Developed animal deterrent systems to protect collection animals from mountain lions without harming any species
- Designed and implemented embedded systems using Arduino for non-lethal wildlife management
- Created circuit designs for solar-powered deterrent systems deployed at conservation facilities
- Utilized 3D modeling for prototyping and optimizing deterrent device housings
- Collaborated with veterinary researchers to ensure humane and effective wildlife protection methods