EECS / NEURO / MATH · UC BERKELEY

Mihir Sharma

I’m a UC Berkeley student studying EECS, neuroscience, and math. I’m really passionate about agentic AI, brain-computer interfaces, robotics, and longevity!

01 //

about

Hi! I’m currently building a startup and I enjoy working on real-world technical problems that help people!

Right now I work with Chang Lab at UCSF on using speech as a biomarker for depression and chronic pain. I also lead the wetware division for Neurotech@Berkeley, working on neurobiologically-inspired reinforcement learning.

Before college, I spent a lot of time in robotics, competitive programming, and debate. Outside of school and startups, I’m usually playing the cello, running, playing basketball, hacking, or going down some rabbit hole about consciousness.

02 //

experience

Co-Founder — Working Memory startup

workingmemory.vip
  • Working Memory aims to become the memory layer for personalized AI, building the structured, high-fidelity workflow data that frontier models need to operate with real continuity across a person's working life.
  • Backed by Mayfield AI Fund, Berkeley SkyDeck Pad-13, Microsoft for Startups, NVIDIA Inception, and Orrick Tech.

ML Speech Analysis — DBS for Depression & Chronic Pain

Chang Lab · UCSF
  • Early-detection treatment modification for patients with chronic pain and major depressive disorder undergoing deep brain stimulation.
  • Building speech as a robust biomarker for major depressive disorder and refractory chronic pain.
  • Predicting symptom severity from 15+ acoustic features with SVMs over linguistic content.

Neurotech@Berkeley

Wetware Engineering Connectomics Lead
  • Spearheaded a project on neurobiologically-inspired reinforcement learning.
  • Previously worked on C. elegans neural circuit simulation.

Robotics & Computer Vision Researcher

The AirLab · CMU Robotics Institute
  • Vision transformer research for robot perception.
  • 440+ hours designing a generalized pre-trained Vision Transformer for dense correspondence.
  • Stripped task-specific heads from UniMatch ViT and pre-trained on random feature-map crop matching.
  • Contributed to UniFlowMatch (pub. 06/10/2025) and Map-It-Anywhere (NeurIPS 2024).
  • Employed PyTorch, WandB, Docker, and Singularity on PSC supercomputing clusters.
03 //

projects

04 //

awards & highlights

HackUTD
1st place grand prize
FIRST Robotics
2nd place all-time world hall of fame team 8393
Lincoln-Douglas Debate
world champion, 1st place gold medalist
USABO
national semifinalist
USACO
gold qualifier
Global Neurohack
3rd place, blue sky track
05 //

get in touch

~/contact — zsh
mihir@berkeley:~$ cat contact.txt
linkedin: in/mihir-sharma
mihir@berkeley:~$