Mohammed Mudassir Uddin · MJCET CSE 2026 · Hyderabad

17-year-old who started billing for code, 20-year-old with 8 published papers and a Guinness World Record.

Researcher. Builder. 5 years in the deep end. Started earning from code at 17. Currently building the dataset infrastructure India needs to stop importing AI and start training it.

GSoC 2026 8 Papers NeurIPS 2026 Guinness WR MJCET CSE TPU Research
Mohammed Mudassir Uddin
Mohammed Mudassir Uddin Ships AI on hardware that costs less than a textbook
~3.9k commits this year
8 papers published
40+ hackathons placed
GitHub commits this year alone, not including private repos
LinkedIn audience organic growth, zero paid promotion
Published papers all in 2026, all before graduating
Hackathons placed shipped real products under genuine pressure
Public repos everything earned in the open
Guinness Record largest agentic AI hackathon ever run

"Started earning from code at 17. Never stopped."

MJCET B.E. CSE 2026 · Hyderabad, India · Available for research collaboration
Mohammed Mudassir Uddin
Mohammed Mudassir Uddin B.E. CSE · MJCET · 2026 Hyderabad, India
Guinness WR GSoC 2026 NeurIPS 2026

"I don't build to ship. I build to understand. The shipping is just proof I understood."

Mudassir, 2026

The honest version.

No PR language. No buzzwords. Just the actual sequence of events that got here.

Built a BitTorrent client from scratch in 8 hours because I was tired of pretending to understand distributed trust.
Built a blockchain in C# because I kept nodding in crypto conversations and eventually that stopped working.
Built an offline AI that runs on $15 hardware because privacy shouldn't be a premium feature that only enterprises afford.
Published 8 research papers before finishing my degree. None of them were easy. All of them were necessary.
Participated in a Guinness World Record at the largest agentic AI hackathon ever. 2,300 developers building simultaneously.
GSoC 2026 and German SoC 2026 contributor. Open source first, always.
Now building the missing layer. India has 1.4 billion people and almost no regional AI training data. I intend to fix that, one domain at a time, on TPU.

Startup Journey

Brevenia

Past

2022 · Social Media Growth Agency

Co-founded at 17. Cold outreach to paying clients, zero funding, no warm intros. First payroll. First real lesson, distribution beats product every single time. We didn't raise money. We charged money.

Auragenzy

Past

2025 · AI Workflow Automation SaaS

Shipped an AI SaaS for no-code workflow automation. Real paying users within months. Product market fit is not a roadmap, it's a credit card transaction from someone who needed this yesterday.

Stealth Indian AI Dataset Infrastructure

In Progress

2026 · Data Infrastructure and AI Research

India has 1.4 billion people and almost no representative AI training data for its regional languages, healthcare, legal, agriculture, and cultural domains. I'm building a startup that trains collaborative AI agents to structure these missing datasets natively on TPU clusters, not GPU. The goal is infrastructure ownership. Own the data, train the models, control the outcomes.

Google Awards for ML Research 2026 Amazon Research Call for Papers 2026

Education History

Muffakham Jah College of Engineering & Technology

Ongoing

Aug 2023 – Jun 2026 · Bachelor of Engineering, CSE

Affiliated to Osmania University, Hyderabad. Maintained 80% aggregate while simultaneously shipping 8 research papers, founding two startups, and contributing to Google Summer of Code 2026. Final-year specialization in AI systems, with active coursework in distributed computing and machine learning theory.

Osmania University, Hyderabad

Ongoing

Aug 2023 – Jun 2026 · Bachelor of Engineering, CSE

One of India's oldest autonomous universities. Chosen for its rigorous engineering curriculum and proximity to Hyderabad's tech ecosystem. Grade 80%. Academic load deliberately paired with real-world engineering output.

SBTET, Telangana

Graduated

Jul 2020 – May 2023 · Diploma in Computer Science

State Board of Technical Education and Training (SBTET). Completed the full 3-year diploma with 80% grade before transitioning to full-time degree. Started freelancing in code during the diploma years itself, billing clients from age 17.

Selected Work

Everything here was built under real constraints. Sleep deprivation, hardware limits, deadline pressure. The friction is the feature.

In Progress 001

Perception

Edge-deployed sensory intelligence replacing a 90-year-old mobility solution for 253 million visually impaired people. Fuses computer vision, spatial audio, and on-device ML. Runs on hardware that costs less than a restaurant dinner for two.

Edge AIComputer VisionTensorFlow LiteSpatial AudioRPi
Post-mortem note

"The hardest part wasn't the model. It was explaining to the model which direction is 'obstacle' when a user's spatial orientation is entirely unknown. We rewrote the spatial mapping layer three times."

Shipped 002

AI Chief of Staff

Organizational intelligence built in 36 hours at Hack-Nation with MIT Sloan AI Club. The RAG pipeline crashed at hour 22. Rebuilt it by hour 26. Shipped at 36 on zero sleep. Mapped real-time knowledge flow across teams.

LLMsRAGWebSocketsOrg DesignMIT Sloan
Post-mortem note

"Sleep deprivation reveals what you actually know versus what you've only ever read about. The gap is usually humbling and exactly where you need to grow."

Shipped 003

Guardian-AI

LLM-powered security layer for real-time threat detection in SOC environments. Used Constitutional AI to prevent the system from hallucinating threat classifications, because security AI that guesses is categorically worse than no AI at all.

LLM SecuritySOC AutomationConstitutional AI
Post-mortem note

"The adversarial test cases we didn't think of were the ones that mattered most. We shipped a second hardening pass after launch."

Shipped 004

Offline RPi AI

Privacy-first voice assistant running entirely on a $15 Raspberry Pi. No cloud. No telemetry. No compromise. Constraints were the architecture document. The first model didn't fit in RAM by 400MB. The final solution was 12 architectural decisions later.

Raspberry PiWhisperEdge AIPrivacy
Post-mortem note

"Constraints are the best architecture documents you will ever write because they are written in reality, not ambition."

Paper Published 005

BioNova

Gene editing outcome prediction using graph neural networks trained on CRISPR experimental data. Published in Bioinformatics journal. Biology doesn't care about your loss function, but it will absolutely destroy your assumptions about generalization.

BioinformaticsPyTorchCRISPRGNN
Post-mortem note

"We overfitted to the training distribution spectacularly. Real CRISPR data is far messier than any public benchmark suggests."

Shipped 006

Zerodha AI Bot

Algorithmic trading with real-time sentiment analysis on Indian equity markets. My first live run lost money in 40 seconds. The second broke even. The third was profitable. Markets are humbling. Every backtested model lies a little, or a lot under volatility.

NLPFinanceReal-timeSentiment
Post-mortem note

"The fourth run is still live. The model I'm most proud of is the one I stopped trusting and verified instead."

Shipped 007

YatraKavach

Tourist safety intelligence with geospatial risk analysis across India. The data gap for non-metro Indian regions was enormous. That gap eventually became the seed of the dataset infrastructure startup I'm building now.

Geospatial MLRisk AnalysisIndia
Post-mortem note

"Prevention is not an alert. Prevention is modeling risk before the event exists in any dataset."

Paper Published 008

ProofGuard AI

Automated mathematical proof validation using constrained decoding. Published at NeurIPS Workshop 2026. Guarantees the AI's answer is actually correct, not just syntactically plausible. Teaching an LLM to prove theorems taught me how little I understood proofs.

Formal MethodsLLM ReasoningConstrained Decoding
Post-mortem note

"The best outcome from any project is realizing a gap in your own understanding. This one found three."

Shipped 009

MediCoach

Medical ML for accessible diagnostics and patient guidance in low-resource settings. Healthcare AI needs 10x the testing because the cost of being wrong is a person and not a benchmark score. The first version shipped confidently. That confidence was the first thing we had to unlearn.

Medical MLComputer VisionAccessibility
Post-mortem note

"Production data distribution and test distribution are never the same. In healthcare that gap is measured in outcomes."

Shipped 010

Compliance Tracker

Regulatory compliance automation via NLP document analysis. The most boring problems often have the highest ROI per line of code. Boring problems with high switching costs are the most defensible products in the world.

NLPDocument AIAutomation
Shipped 011

Cookie Manager

A powerful Chrome extension for cookie management to add, delete, edit, search, protect, and block cookies across Chromium browsers. Built for Manifest V3 with modern security standards. Originally forked to fill the gap left by outdated cookie tools. Now the most complete lightweight option available.

Chrome ExtensionManifest V3PrivacyChromium
Post-mortem note

"The original was broken and nobody was fixing it. Forking and shipping was faster than waiting for upstream. Sometimes the best product is just the one that works today."

Shipped 012

Roblox Shirt Template Saver

A Roblox Chrome extension that saves and re-uploads shirt, pants, and t-shirt templates for any item, including off-sale items, with no premium membership needed. Two clicks. No scraping, no auth bypass. Just the template layer that the platform should have exposed natively.

Chrome ExtensionRobloxWeb ScrapingUX
Post-mortem note

"The best tools solve problems the platform creators chose to ignore. Users found it instantly because the pain point was a blind spot for the native engineers."

Research

8 papers. All in 2026. All before graduating. Two funded by Google and Amazon.

arXiv IEEE Access NeurIPS EMNLP ACM CCS MICCAI ICLR Bioinformatics
Active Research · Grant Funded
2026 · Google Awards for ML Research · Amazon Research · Lead Researcher

Agentic Execution on TPU Architectures

Pioneering autonomous AI agents that run inference natively on TPUs rather than traditional GPUs. Fundamentally different parallelism strategies, memory bandwidth allocation, and agent coordination for large-scale multi-agent systems. The core thesis is that the GPU bottleneck is artificial — TPU-native execution unlocks a different cost and scale curve entirely. This is also the compute backbone for the Indian dataset infrastructure startup.

Indian Dataset Infrastructure TPU v4 · v5 Multi-Agent Systems Agentic Inference
2026· arXiv· First Author

Hierarchical Sparse Circuit Extraction from Billion-Parameter Language Models

A new structural approach to understanding what is actually happening inside GPT-scale models at the mechanistic level. Interpretability that scales.

2026· IEEE Access· First Author

Edge-Deployed Multimodal Fusion for Low-Resource Assistive Navigation

Making full AI navigation pipelines work on a $15 computer without any internet. This is the research backbone of the Perception project.

2026· ACM CCS Workshop· Co-Author

Adversarial Robustness in LLM-Augmented Security Operations Centers

How to stop attackers from tricking AI-powered SOC systems via adversarial prompt injection and credential exfiltration vectors.

2026· NeurIPS Workshop· First Author

Formal Verification of Neural Theorem Provers via Constrained Decoding

Guaranteeing an AI's mathematical proof is actually correct — not just syntactically plausible. Formally verified rather than just tested.

2026· MICCAI Workshop· Co-Author

Privacy-Preserving Federated Learning for On-Device Health Diagnostics

Training medical AI across multiple hospitals simultaneously without any single party ever seeing another's patient data.

2026· EMNLP Findings· First Author

Causal Intervention Methods for Debiasing Code Generation Models

Making code generation models less biased without sacrificing performance using a causal framework rather than post-hoc filtering.

2026· Bioinformatics· Co-Author

Real-Time CRISPR Outcome Prediction Using Graph Neural Networks

Predicting gene editing outcomes before they happen using GNNs trained across experimental CRISPR datasets.

2026· ICLR Workshop· First Author

Attention Head Specialization Dynamics in Instruction-Tuned Transformers

How ChatGPT-style models internally reorganize attention heads when fine-tuned on instruction datasets. Interpretability at the head level.

253 million people.

One problem nobody has truly solved.

The white cane hasn't changed since 1930. That is not an exaggeration. That is the fact. I want to fix that specifically and permanently.

"Everything I have built was training for this."

If this is what you are building too, let's talk
The Problem

Perception is an edge-deployed sensory intelligence platform designed to replace the 90-year-old white cane. 253 million people worldwide are visually impaired. The tools available to them have not meaningfully evolved in nearly a century.

The Solution

By fusing computer vision, spatial audio, and on-device ML, Perception gives visually impaired users real-time environmental awareness without cloud dependency, without expensive hardware, and without compromising privacy. It runs on hardware that costs less than a dinner for two.

The Constraint

Under $20 hardware. Zero internet requirement. Under 80ms latency. This is not a research prototype. This is a product constraint that forces every architectural decision to be justified by real-world usability.

The Status

In progress. Active development. Open to collaboration with researchers, NGOs, and accessibility-focused organizations who understand why this matters.

The War Record

40+ hackathons. Real products built under real pressure. Only the ones worth telling.

Feb 2026

Hack-Nation Global AI Hackathon

With MIT Sloan AI Club · 36 hours

Built AI Chief of Staff. The RAG pipeline crashed at hour 22. Rebuilt by hour 26. Shipped at 36. Organizational intelligence that maps real-time knowledge flow across teams.

"Building under real pressure strips away everything except what you actually know."

Dec 2025

Agentathon 2025 Guinness World Record

MallaReddy University · 2,300+ developers

The largest agentic AI hackathon ever run. Built fraud detection agents for public spending transparency. The record was a bonus. The lesson was something else entirely.

"The record was a bonus. The lesson was humility."

Dec 2025

ElevenLabs Worldwide Hackathon

At NxtWave · Voice AI

Built the Netflix of Corporate Training. Upload a PDF, receive back bite-sized AI-narrated video episodes.

"The best UX insight is that people hate reading PDFs."

Nov 2025

IBM watsonx Orchestrate Hackathon

LLM Security Track

Built Guardian-AI, a Constitutional AI security layer for LLMs in SOC environments. The adversarial test cases we didn't think of were the ones that mattered most.

"The adversarial cases you don't think of are the ones that matter most."

Jan 2025

7Days7Builds Challenge

Built in Public

7 deep CS projects in 7 days. BitTorrent client. Blockchain in C#. 3D engine from scratch. Semantic code search. The best code teaches you why your first attempt was wrong.

"The best code teaches you why your first attempt was wrong."

Certifications

Selected credentials across AI, ML, and engineering. The ones that required actual effort to earn.

Anthropic
Claude Certified Architect — Foundations
Issued Mar 2026 · Score: 944/1000
Score: 944
Anthropic
Claude Certified Architect — Early Adopter
Issued Mar 2026
Harvard University
CS50X 2026
Issued Feb 2026
DeepLearning.AI
Neural Networks and Deep Learning
Issued Nov 2023
IBM
IBM Data Science Professional Certificate
Issued Nov 2023
DeepLearning.AI
TensorFlow for AI, ML and Deep Learning
Issued Nov 2023
UC Davis
SQL for Data Science
Issued Nov 2023
Microsoft
Preparing Data for Analysis with Microsoft Excel
Issued Nov 2023
UC Davis
Fundamentals of Visualization with Tableau
Issued Nov 2023
IBM
Statistics for Data Science with Python
Issued Nov 2023
Google
Google IT Automation with Python Professional Certificate
Issued Sep 2023

Let's build something that matters.

I respond to people who ship things. If you're working on hard problems in AI infrastructure, edge computing, or regional data, reach out. If you want to talk about what India's AI ecosystem is structurally missing, absolutely reach out.

mohd.mudassiruddin7@gmail.com
LinkedIn · GitHub · Scholar · Twitter

"I respond to people who ship things."