Computer Science student with research experience in LLM evaluation, data analysis, and ML-based systems. NeurIPS 2025 co-author. Authorized to work in the US on OPT/CPT.
I work on the layer where research meets product, designing benchmarks, evaluation pipelines, and ML-powered tools that turn raw model capability into something useful and measurable.
Recent work spans a multi-LLM benchmarking platform at UC Davis, a NeurIPS 2025 benchmark on adversarial bias detection, and shipping production AI dashboards with Claude, Gmail, and Drive APIs. I move between Python, TypeScript, and Swift, and care equally about model behavior and the seams between systems.
Originally from Madagascar, I pursued my studies in International Business in England, where I immersed myself in diverse cultures and gained a deep appreciation for global markets.
Eager to expand my skill set, I am now advancing my expertise in Computer Science in the United States, merging my passion for technology with the strategic perspective I acquired through my business background.
iOS app turning the iPhone into a smart dashcam. Rolling buffer saves the last 60s of dual-camera footage to MongoDB. Real-time perception pipeline streams frames to a FastAPI sidecar running YOLOv8, fused with ARKit LiDAR scene-depth for sub-100ms hazard scoring on close passes, doorings, and blocked bike lanes.
Deployed AI automation dashboard for a client company. Auto-generates daily business briefings from Google Drive and schedules AI-drafted client emails via Gmail. Human-in-the-loop outbox with one-click delivery. OAuth tokens secured with AES-256-GCM on a Prisma/Postgres backend.
Offline iOS app converting field worker voice notes into structured agronomic observations via on-device LLM inference in under 90 seconds. Local rules engine generates time-bounded treatment recommendations from live weather features, with playbook patching and version-tracked audit trails.
Sports app letting students find pickup buddies and meetups nearby. Full-stack on Google App Engine with a profile-based AI recommendation engine. 20+ active users.
MultiAgent Diplomacy — an AI strategy game where LLM agents autonomously negotiate and compete via Langchain. Integrated Google Cloud TTS to synthesize natural voice between agents.
Real-time wildfire detection alerting California residents to active fire threats via live camera monitoring. ML model trained on 21,000+ images detects smoke and fire across 1,150 traffic cameras. Predictive risk model trained on 14,000 historical incidents.
Coursework: Software Engineering, Artificial Intelligence, Computer Architecture, Algorithm Design & Analysis, Programming Languages, Theory of Computation, Operating Systems. Active in the Google Developer Student Club, ML lab research, and hackathons.
Coursework: Data Structures & Algorithms, Object Oriented Analysis & Design, Linear Algebra, Discrete Mathematics. Director's Choice Award at De Anza Hacks 2.5 for an audio-reactive LED & haptic visualization device.