Vedant Gupta

I’m building AI that reasons reliably at scale. To me, this problem combines the engineering thrill of creating systems that work (imagine AI writing proofs and coherent codebases!) with the scientific joy of understanding what intelligence fundamentally requires.

I am currently a Research Engineer at Harmonic. Previously, I was a Founding Member of Technical Staff at Asari AI, where I built core infrastructure and performed applied research to create reliable AI agents that perform systems-level reasoning, backed by Eric Schmidt, Jeff Dean, Caltech, etc.

I recently graduated from Brown University with a B.Sc. in Mathematics and Computer Science (with university and departmental honors). Alongside my research, I captained the Brown Men’s Squash Team and interned at the Boston Dynamics AI Institute (now RAI Institute).

I’ve worked at the experimental and theoretical ends of CS research. On the experimental side, I worked at the Intelligent Robot Lab with George Konidaris to build algorithms that improve robot generalization outside their training distribution. On the theoretical side, I worked with Lorenzo De Stefani to analyze the I/O complexity of Dynamic Programming algorithms, aiming to make algorithms run faster on hierarchical memory systems.

Learning Parameterized Skills from Demonstrations
Neural Information Processing Systems (NeurIPS) 2025
V. Gupta, H. Fu, C. Luo, Y. Jiang, G. Konidaris
[paper] [github] [website]

An algorithm that learns interpretable parameterized skills that improve generalisation to unseen tasks, especially in data-constrained settings and on out-of-distribution tasks.

On the I/O Complexity of the CYK Algorithm and of a Family of Related DP Algorithms
Algorithms and Data Structures Symposium (WADS) 2025
L. De Stefani and V. Gupta (equal contribution)
[paper]

Introduces tight bounds on the number of I/O operations required when running key DP algorithms on hierarchical memories.

Robot Task Planning under Local Observability
IEEE International Conference on Robotics and Automation (ICRA) 2024
M. Merlin, S. Parr, N. Parikh, S. Orozco, V. Gupta, E. Rosen, G. Konidaris
[paper]

Synthesizing Navigation Abstractions for Planning with Portable Manipulation Skills
Conference on Robot Learning (CoRL) 2023
E. Rosen, S. James, S. Orozco, V. Gupta, M. Merlin, S. Tellex, G. Konidaris
[paper]

A sampling of smaller scale projects and side quests.

TinyLM: How small can language models get while still producing coherent sentences? [code]

K9DB: An open source database system for GDPR compliance-by-construction. I contributed optimizations that halved memory overheads. Published at OSDI ‘23. [code]

Free Will Lean: A formal proof of the Kochen-Specker paradox using Lean 4. [code]

Spot Tour Guide: An interactive robot tour guide that interacts with users while generating custom tours on the fly. Built at the RAI Institute.

Fixing Option Critic: Explored complex solutions to prevent option-critic algorithm from suffering options collapse. Then realized you only need to add an extra hyperparameter in the loss function. [writeup] [code]