About
fourth year cs & ai, rank 1 in my cohort. ← i mention this unprompted.
taught a neural network that electricity has to obey kirchhoff’s laws. it refused. IEEE accepted us anyway.
founding engineer at a dubai startup in my second year. contribution margin up 60%. they raised a series a. i was, functionally, moving boxes around in a database.
spent three weeks on a puzzle contest to get into a hackathon. did it again the next year. 17th in the world, then 8th. something is wrong with me :)
anyway. scroll ↓
Work Experience
SanasOffer received
Transient AI Inc
Zuvees
Zota Health Care Ltd
Education
Newton School of Technology, Rishihood University
Honors & Awards
HackMIT 2026 — 8th Globally
September 2026HackMIT · Puzzle Solver Contest · Meta & Visa challenges at MIT, shortlisted among 25 of 200 teams in the first round
IEEE PES Energy Shark Tank 2026 — Top 5 & 3rd Place
August 2026IEEE PES Young Professionals · APPEEC Singapore · FlexGrid pitch · Certificate of Achievement
University Topper — B.Tech (CS&AI)
2026Newton School of Technology, Rishihood University · Rank 1 in cohort as of third year
Y Combinator Startup School India 2026
2026Y Combinator · Invite-only founder program, Bangalore
1st Place — NST Startup Foundry 2026
2026Jarvis, an AI Personal Intelligence System · pitched to Google Cloud & Microsoft for Startups
Google Big Code Challenge — National Top 50
2026Google · Invited to the 48-hour Challenge Finale · awarded a Google Pixel
Berkeley Global Access Program
2026UC Berkeley · Selected for the Spring 2026 and Fall 2026 semesters
Runner-Up — Voloridge Sponsorship Challenge
September 2025Voloridge × HackMIT · HackMIT 2025 · 17th globally in the Puzzle Solver Contest
Pre-Placement Offer
2025Zuvees · Built a first-of-its-kind Delivery Slot Management System
JEE Advanced 2023 — All India Rank 5649
June 2023National Testing Agency · Offered a CSE seat at IIT Goa
NTSE Scholar
January 2020NCERT · National Talent Search Examination
2nd Place — State-Level Public Speaking Contest
The Hindu
Skills & Tech Stack
Check out my latest work
I've worked on a variety of projects, from compilers and CUDA kernels to research models and full-stack products. Here are a few of my favorites.
Turtle Compiler
A compiler you can watch work. Type a turtle-graphics program and all six phases run live in the browser, from tokens and an LL(1) parse trace to three-address code, six optimization rules and bytecode for a small VM that draws the picture. After optimizing, it runs both versions on the VM to confirm the picture is unchanged. Built with a team of five for the Compiler Design and Automata course.
Does a 98%-accurate malaria detector still work on patients it has never seen? We rebuilt a published EfficientNet-B0 ensemble from its authors' own code, matched it within 1 pp on every headline metric and found two errors in the paper. Then, in a pre-registered test, the ensemble scored 98.45% on 40 unseen patients. Written up as an 8-page IEEE-format paper. Team of four; I built the reproduction pipeline and ran the GPU training.
Labels every exam question with its Bloom's taxonomy level, then rewrites questions until the paper meets its blueprint. Tested on question sources it never trained on (9,855 questions from 7 sources), it beats the standard TF-IDF + SVM by +0.051 weighted F1 (p < 0.001). Colleges label the same verb differently ("Explain" is Remember in 55% of one set and Understand in 93% of another), so a college can teach it its own convention with a few labelled questions. A pipeline of agents plans, rewrites and verifies each question, all on a CPU.
cuCopula
A GPU engine for tail-risk scenarios: numerically stable CUDA kernels for a parallel copula method, with the tests a risk desk would ask for. On an H100, covariance runs 25x faster end to end than a 22-thread CPU, and sampling 18x. A pre-registered 99% VaR backtest on 50 large US stocks (2015 to 2025) showed that volatility filtering matters far more than the choice of copula.
SIEVE
Pattern mining that stays honest on autocorrelated time series. It tests tens of thousands of if-then rules and keeps only those that survive a block-permutation correction. On data with no signal, it raises a false alarm 4% of the time; the standard row-shuffle correction, 100%. Given 36 years of VIX, it rediscovered volatility clustering on its own. Custom CUDA kernels run 112x faster end to end than single-core NumPy. Started solo at HackMIT 2026.
Spotify Wrapped for your friends' shopping, built at HackMIT 2026 with a four-person team; I co-built the Fastify + TypeScript backend. Taste match is computed from real purchases, Claude picks gifts that cite items the recipient saved, and Visa Direct splits the payment. Shortlisted among 25 of 200 teams in the first round of the Meta and Visa challenges.
Implemented floor vs. non-floor pixel/region classification on indoor images using the CMM dataset and Support Vector Machines. Compared three feature-engineering approaches: RGB-only pixels, RGB plus spatial coordinates, and KMeans-based region-level features. Method 3 (KMeans regions) achieved ~92.5% test accuracy with the fastest training.
Watt-IF
Physics-informed forecasting for power grids. Stage 1, KAT-PatchTST, is my first-author paper at IEEE APPEEC 2026, published in IEEE Xplore: PatchTST with a Kirchhoff conservation regularizer reaches 3.65% demand MAPE across six US balancing authorities, within 0.78 pp of the best weather-integrated baseline. Stages 2 and 3 (grid bottleneck analysis and a grid-partition policy) are in development, targeting ICML.
I like hacking things (a lot)
During my time in university, I attended 6+ hackathons. It was eye-opening to see the endless possibilities brought to life by a group of motivated and passionate individuals within 2-3 days. I have made some of my best friends and memories at these hackathons :)
- H
HackMIT 2026 (8th Place Globally 🏆)
CodeLive demoCambridge, MAAchieved 8th place globally in the HackMIT Puzzle Solver Contest 2026, improving on 17th place in 2025. Participated in the Meta and Visa sponsor challenges at MIT with Unwrap, built with a four-person team (MIT, Oxford, Vanderbilt, NST); shortlisted among 25 of 200 teams in the first round. - G
Google Big Code Challenge
Top 50 nationally in the Google Big Code Challenge; invited to the 48-hour Challenge Finale and awarded a Google Pixel device. - H
HackMIT 2025 (17th Place Globally 🏆)
Cambridge, MAAchieved 17th place globally in the HackMIT Puzzle Solver Contest 2025 and runner-up in the Voloridge sponsorship challenge. Built a distributed log analyzer using parallel computing. - I
IEEEXtreme 18.0 (2024)
Improved to a global rank of 775 and All-India rank of 224 in IEEE Hackathon. - M
Meta Hacker Cup 2024
Secured rank 4553 in Meta Hacker Cup 2024. - I
IEEEXtreme 17.0 (2023)
Achieved a global rank of 1305 and All-India rank of 446 in IEEE Hackathon.
Get in Touch
Want to chat? Feel free to reach out via LinkedIn or email me directly at cmurali.m23csai@nst.rishihood.edu.in.