Manish Ram
Hi, I'm Manish. I co-founded Parafield and I'm its Chief Scientist. We make spatial audio for audiobooks and audio stories. I also work on machine learning for health: a foundation model for fMRI, DeepTracer's protein diffusion model, medical language models, and the benchmarks and RL environments we test them with.
About
What I've been working on
ICML & NeurIPS papers
Co-author of two ICML 2026 papers and one at NeurIPS 2026. The University of Washington gave me its Undergraduate Research Excellence Award for reaching ICML before graduating.
CortexMAE fMRI model
Contributed to CortexMAE, an fMRI foundation model on Vision Transformers that decodes cognitive states in real time.
Medical LLM RL environment
Built an RL environment on Prime Intellect that cut medical LLM hallucinations by 20–24%, by checking every claim in an answer.
Kosmos for narrators
Kosmos: free, open-source software for narrators, with mastering, a teleprompter, and spelling and pronunciation checks. Our first customer uses it.
Protein diffusion model
Led training and evaluation of DeepTracer's protein diffusion model (IEEE BIBM 2026 workshop), and got UW's Undergraduate Leadership and Service Award for it.
Open-source contributions
Pull requests to Swift, Turborepo, Vercel's Chat SDK, Backstage, NASA's F´ and AWS Powertools.
Early recognition
In grade 9, part of a Guinness World Record: the most people taking an online programming lesson in 24 hours. Top 5% in India's national STEM boards, twice.
Papers
Published research
CortexMAE at ICML 2026
Scaling Vision Transformers for Functional MRI with Flat Maps. arXiv 2510.13768.
Medmarks at ICML 2026
An open-source LLM benchmark suite for medical tasks. arXiv 2605.01417.
Writing
Recent pieces
Parafield vs. Dolby Atmos
How the Parafield Spatial Engine Compares With Dolby Atmos. Oct 2026 · Parafield, with Thomas Suen and Justin Liao.
Volume isn't distance
Why Turning a Sound Down Doesn't Make It Sound Far Away. Oct 2026 · Parafield.
How two ears help
How Your Two Ears Keep the Narrator Clear. Sep 2026 · Parafield.
Sound inside your head
Why Headphone Sound Sits Inside Your Head. Aug 2026 · Parafield.
Two Minds, One Model
Notes on HealthGPT. Jun 2026.
Building a medical RL env
Building an RL Environment for Medical QA. May 2026.
Grand Rounds for Machines
Three medical benchmarks. Mar 2026.
Reading brain scans
Teaching a Model to Read Brain Scans. Feb 2026 · CortexMAE, ICML 2026.
Tools I use
Python & PyTorch
Core stack for model training, evaluation, and research.
scikit-learn & C++
Classical ML and performance-critical systems work.
TypeScript & React
Frontend and full-stack development, including Node.js and Vite.
Swift, Dart & Flutter
Native and cross-platform mobile apps.
AWS & Docker
Cloud infrastructure and containerised deployments.
Git & GitHub Actions
Version control and CI/CD across every project.
Linux & Bash
Day-to-day environment for scripting and server work.
LaTeX
Writing and typesetting research papers.