Work experience

TODO – TODO

TODO

Built agentic security-analysis tooling on top of an MCP server, connecting retrieval and specialized tools for network and system investigation workflows.

• –Designed and implemented tool-oriented agent workflows for security and network analysis questions, backed by an MCP server
• –Built retrieval and context-grounding components so agent responses were backed by real telemetry rather than speculation
• –Wrote tests for multi-step agent workflows and reasoned about reliability and failure modes
• –Worked within existing container/orchestration infrastructure for deployment

• Python
• FastAPI
• MCP
• PyTorch
• PostgreSQL
• Docker
• Kubernetes
• OpenShift

TODO – Present

New York, NY

Researching efficient inference for language and vision-language models, focused on compression methods that preserve task quality.

• –Implemented low-rank and activation-aware compression methods for language and vision-language models
• –Built an evaluation harness to compare compressed and baseline models across memory, speed, and task quality
• –Ran experiments on Slurm-scheduled GPU compute and iterated on reproducible benchmarking practices

• Python
• PyTorch
• CUDA
• Slurm
• Hugging Face

TODO – TODO

TODO

Built web platform features for online classrooms and educational workflows.

• –Developed features for the online classroom platform used by students and instructors
• –Worked across the frontend and Firebase/Firestore-backed data layer to ship classroom workflow features

• React
• JavaScript
• TypeScript
• Firebase
• Firestore