Emora Health, AI engineering intern. Mental-health screening and referral workflows. Built an agentic video pipeline that took performance-ad generation from $9–13/min to under $1.50, and a referral-profiling pipeline that turned 50+ hours a week of manual research into under 2.
Udit Akhouri
Hi. I'm a final-year CS student at IIT Patna and an independent ML researcher. I spend most of my time asking one question: can thinking differently make machines reason better? This summer I was an AI engineering intern at Emora Health, a New York based paediatric mental health company.
What I'm working on
I'm interested in whether alternative cognitive strategies, the kind you find in neurodivergent minds, give a computational edge in machine reasoning: novelty seeking, spotting traps, exploring instead of exploiting, generating hypotheses. I turn those hunches into systems people can actually run, plus benchmarks so we can tell if they work.
Broadly: machine reasoning, cognitive architectures, agentic AI, scientific discovery, and how humans and AI make decisions together.
Things I've built
ADHD · a neurodivergence-inspired reasoning harness
What happens if you let an LLM's attention wander on purpose? Divergent search instead of one tidy chain of thought. On my benchmark suite it got ~5.2× better at trap detection and ~2.9× at novelty. 4,000+ stars, 100k+ downloads, adopted by 20+ open-source projects, 10+ contributors. My favorite thing I've made.
NeuroArxiv · STEM reasoning & research engine
ADHD, but it reads the literature first. Pulls from arXiv inside the reasoning loop so the model checks what already exists before proposing something new. 400+ stars, 10k+ downloads.
CDAF · cached descriptive asset files
An open spec so agents stop re-watching the same video over and over. Spec, CLI, agent skill, and a reproducible benchmark. Cut token usage 91% and latency 65%.
Where I've been
May-Sept, 2026
Jan-April, 2026
Incubation Centre, IIT Patna, AI research intern. Runtime compliance infrastructure so healthcare AI workflows leave an auditable trail.
2025–26
Exthalpy, founder. Tried to represent how experts actually make decisions, computationally. $25k in grants and incubation from MeitY (Government of India) via DRISHTI CPS, IIT Indore.
Jan-Oct, 2024
Malpani Ventures, builder in residence. AI modeled on how doctors consult. 4,500+ patient interactions, 700+ consultation hours saved. This is where the tacit-expertise itch started.
2023–24
Sttabot AI, founder. No-code AI platform. Raised ₹30L+, 12,000+ users and ₹24L ARR in 10 months. My first company; I learned a lot.
Misc
- Research funded by the Ministry of Electronics & IT, Government of India.
- Cover story in The New Stack.
- Gave a guest lecture on LLMs and MSMEs at an Asia Pacific Institute of Technology conference, and once talked to a room of BBA freshers about startups.
- B.S. Computer Science, IIT Patna, 2023–2027. CGPA 8.59.