Building systems that replace manual work with intelligence.
MS CS @ USC
Previously @ Google, Zanskar
Most ML work today is bottlenecked not by models, but by systems around them.
I’m focused on:
- turning research into usable infrastructure
- reducing human-in-the-loop workflows to near zero
- building AI systems that operate, not just predict
Upload a research paper → get runnable code
- Parses papers → extracts methods → generates PyTorch repos
- Handles missing details with structured reasoning + retrieval
- Used by early users to cut reproduction time from hours to minutes
Deciding when models should trust themselves vs ask for help
- Uncertainty-aware routing using entropy + margin signals
- Avoids unnecessary RAG while improving accuracy
- Focus: latency, cost, reliability tradeoffs
Long-horizon agent planning
- Skill-based abstraction over low-level actions
- Tech-tree guided exploration
- Combines RL + sequence modeling
Core: Python, Go, Java
ML: PyTorch, Transformers, RAG, LangGraph
Systems: FastAPI, gRPC, Distributed Systems
Infra: Postgres, Redis, Elasticsearch
Cloud: AWS, GCP, Docker
- If it doesn’t reduce real-world effort, it’s noise
- Latency and reliability matter more than marginal accuracy
- Systems > models
- Build things people can’t easily replace
- Email: ved29022004@gmail.com
- LinkedIn: https://www.linkedin.com/in/ved-chadderwala-196529223/
- Portfolio: https://vedchadderwala.com