Yeshasvi Kamma
currently in BerkeleyJunior at UC Berkeley studying computer science and applied mathematics. I'm interested in foundation models, agentic systems, and applied ML for scientific data.
Splitting time between two research roles.
At Cohere Labs, I'm building a vision-language agent for street-level service discovery over Google Street View imagery.
Through Berkeley's Data Discovery, I lead two projects at NASA Ames on the OSDR/GeneLab spaceflight biology corpus: a Metadata Intelligence Engine for pulling verified, ML-ready data, and a Foundation Model for gene expression across tissues and organisms.
Recently joined the OSDR/GeneLab Active Working Group to connect both into a full training and fine-tuning pipeline.
Previously, I led the AI platform at YottaSecure, where I architected YSecure, a serverless AWS Bedrock platform for security intelligence. I migrated the original Rasa + Trivy + OpenAI chatbot into a Bedrock RAG architecture for CVE triage and compliance mapping.
Earlier, I did ML research at NEBDHub at Columbia, training a model to predict NYC crash risk from spatiotemporal features.
A couple of side projects.
Anomaly Intelligence is a real-time news pipeline that surfaces emerging stories through clustering and anomaly scoring.
DeepScout is a multi-agent research synthesizer that turns papers into a knowledge graph with contradiction edges.
Outside research, I train kickboxing and jiu-jitsu in Berkeley. Before that, eight years of Taekwondo and wrestling through high school.
I'm a big Warriors and UFC fan. My favorite movies are Interstellar, Good Will Hunting, Fight Club, and Kill Bill. I listen to Joe Rogan, Lex Fridman, and Theo Von.