About me
Engineer, researcher, pianist, occasional philosopher.
Greetings! I’m Agastya Seth, a Staff AI Engineer at Synopsys in Sunnyvale, CA, where I’m the tech lead for agentic AI in the HAV R&D group. I build AI agents that do real engineering work: planning, running, and debugging hardware verification flows that can take 20 to 30 hours per run.
At Synopsys I lead the agent work behind the customer-facing AgentEngineer and Autopilot products. That means a long-running autonomous orchestrator that dispatches and monitors task agents across the compute farm, an agentic chat engine used by roughly 800 engineers a month, and the security guardrails (scoped data access, sandboxed tool execution, prompt-injection defenses) that make it safe to point agents at proprietary source code and design data. I also own our agent evaluation strategy, because an agent you can’t measure is an agent you can’t trust.
Before that, I spent three years at Cadence Design Systems writing high-performance C++ graph and geometry algorithms for parasitic extraction, and later interned on the Cerebrus team predicting timing and power directly from RTL. Along the way I built an automated essay scoring system for publishers like Wiley at LiqVid, and trained fashion-attribute models at ViSenze in Singapore.
I did my M.S. in Computer Science at Arizona State University (4.0 GPA), as a Graduate Research Assistant in the ACME Lab, after a B.Tech. in Electronics and Communication from Shiv Nadar University. My research sits where LLMs meet safety and hardware: multi-agent LLMs for hardware vulnerability generation (ICCD 2025), agentic test generation (DATE 2026), time-sensitive QA beyond memorization (ACL 2025), and evaluating LLM defense strategies (Findings of ACL 2024).
My passion for music remains a constant companion. As an avid pianist, the art of improvisation and composition continues to be a source of joy and creativity in my life.
Current Focus and Interests
- Agentic systems and multi-agent orchestration
- LLM evaluation and benchmark design
- AI safety, sandboxing, and prompt-injection defense
- AI infrastructure for long-running agents
- LLMs for hardware design and verification
- Reinforcement learning for LLMs
Publications
See the publications list on the homepage, or grab the resume.
Musings
- Edge Computing
- Technology Entrepreneurship
- Sustainable Development
- Music Composition
Through this blog I share projects, experiments, and the occasional wander into the philosophy of life and technology.
I have a million ideas. They all point to certain death.
Marvin, the Paranoid Android
So long, and thanks for all the fish!
Remember, the universe is vast, and the possibilities are endless. Here’s to exploring them together, and yes, sometimes the answer might just be as intriguing as 42.14378291…