About
I build AI systems for places where you can't afford to be wrong. That means enterprise AI at Phyvant and robotics research at Purdue.
I like building infrastructure that works under constraints. Private, accurate, and fast — when conventional wisdom says pick two.
Long-term, I want to build category-defining vertical models that outperform general-purpose AI where precision, trust, and domain context are non-negotiable.
Technical contributions
- Founder of Phyvant — specialized enterprise AI with verified organizational context.
- Helped build RLHF and model quality systems at Mercor.
- First author, PathCluster: Pedestrian Group-Adaptive Social Navigation in Dense Crowds — IROS 2025 (with Aniket Bera).
- MIRACLE: Inverse Reinforcement and Curriculum Learning Model for Human-inspired Mobile Robot Navigation — arXiv 2023 (with Kshitij Tiwari and Aniket Bera).
Interests
Privacy-preserving ML. RLHF (PPO, DPO). Data curation for specialized models — less data, higher signal. Robotics and navigation: human-aware planning and control under uncertainty. Distributed systems that don't go down.
I also think a lot about the business side of enterprise software. Longer cycles, more stakeholders, higher trust requirements. Frustrating and fascinating.
Non-professional interests
- Chess (competitive play and puzzles)
- Basketball (playing and watching)
- Snowboarding (winter slopes and powder)
- Hiking (trying to do more of this)
- Origami (started as a focus thing, now I just like it)
- Anime and manga (Steins Gate, Vinland Saga, Monster)
- Meeting people building things that matter
Contact
nihalgunukula@gmail.com. I read everything.