TANMAY.EXE / A LITTLE WORLD
A little world, getting ready…Scheduled routine · IST
Home / Navi Mumbai
An isometric pixel apartment with a bedroom, study desk, green sofa, books, plants and a view of Navi Mumbai.
Hover over me.
I might have something to say.Tap the character.
He notices.
Next: Sleep at 06:30 ISTA little illustration of my routine.
02:30 IST

projects.exe / selected work

FILEEDITVIEWC:\TANMAY\PROJECTS
04 PROJECTS / OPEN FOR INSPECTION

Things I’ve built.

The code, the decisions,
and what I learned along the way.

Showing all four projects

01 / AGENT EVALUATION

The Tireless Hand.

An agent injects realistic bugs into a drone cockpit. Independent checks compare the UI against telemetry.

11/13 injected bugs caught · 7 checks
GITHUB ↗
02 / HUMAN–AGENT COLLABORATION

Driving Seat.

A Claude Code skill that keeps the developer in charge of intent, tradeoffs, and consequential decisions.

Human judgment stays in the loop.
GITHUB ↗
03 / REPUTATION & AI PERCEPTION

Vigil.

A product I built to understand a brand’s online reputation—and how AI models describe it.

From NeuroChain prototype to a product.
VISIT VIGIL ↗
04 / VIDEO AUTOMATION

Reel-MMA.

Motion, audio, and vision analysis turn MMA training footage into candidate 9:16 highlights for human review.

Experiment paused because of API costs.
GITHUB ↗
More in my GitHub folder ↗
4 itemsBUILD. TEST. LEARN.

project_notes.txt

MY ROLE
BUILT WITH

THE IDEA

HOW IT WORKS

VIEW PROJECT ↗READ MORE ↗
Project notesNO MAGIC. JUST ENGINEERING.

open_source / contributions.log

PATCHES, FIXES & THINGS THAT MATTER

Better, together.

Useful changes to tools
other people build with.

Status checked 22 Oct 2026ALL CONTRIBUTIONS ↗
5 selected contributionsOPEN SOURCE, OPEN NOTEBOOK.

readme.txt / the human

NAVI MUMBAI, INDIA / HELLO, WORLD.

I like knowing
how things work.

And then finding out where they don’t.

I’m Tanmay, an AI engineer studying Artificial Intelligence & Data Science at Terna Engineering College, University of Mumbai.

I build products, agents, and evaluation systems. My favorite part is turning an idea into something useful. A close second is taking it apart until I understand why it fails.

Currently working on agents and evaluation, learning GPU kernels, and going down the occasional robotics rabbit hole.

Off the keyboard? Usually on the mats. BJJ and MMA are part of my life—and the reason Reel-MMA exists.

EXPERIENCE / LEARNING BY DOING

AUG 2026 — NOW
Forward Deployed Engineer
Navayuvak Entrepreneurs

Designing and building founder-assessment infrastructure, mobile-first intake, and longitudinal intelligence.

OCT 2025 — NOW
Founder & Builder
Vigil

Built a brand-reputation and AI-perception product, with feedback from PR and GEO agencies shaping the work.

MAR — MAY 2026
AI Engineer Intern
HaTa Bevtech · Hugg Beverages

Booking and CRM systems, a buying-intent classifier, and production automation for customer support and advertising.

DOWNLOAD RÉSUMÉ ↓
A person, not just a stack.STILL FIGURING THINGS OUT.

research / execution_divergence.txt

LLM EVALUATION / ONGOING RESEARCH

Explanation ≠ execution.

Does being able to explain a formal rule mean a model can reliably carry it out?

DESIGN
900 preregistered instances
MODELS
Llama 3.1 & Gemini 2.0 Flash

THE EXPERIMENT

Three complexity levels and three token distributions, comparing formal-rule explanations with actual rule execution.

THE PART I TAKE SERIOUSLY

An annotation audit found 45.5% inter-rater agreement. I demoted the headline result and rebuilt the pipeline instead of treating an unreliable measurement as a conclusion.

Submitted to ACL Rolling Review, targeting BlackboxNLP 2026. This is ongoing research, not an accepted-publication claim.
LET’S TALK RESEARCH ↗

contact.exe / new conversation

TO: TANMAY SURTE

Got something
in mind?

A project, an opportunity, or a good technical rabbit hole. I’d love to hear about it.

OPEN YOUR EMAIL APP
GITHUB ↗LINKEDIN ↗RÉSUMÉ ↓
No forms. Just a conversation.:)

toolbox.txt

LANGUAGES
Python, TypeScript, JavaScript, SQL
PRODUCT
Next.js, React, FastAPI, Node.js, Supabase
AI & EVALUATION
LLM agents, RAG, PyTorch, scikit-learn, mutation testing
AUTOMATION & INFRA
Playwright, n8n, Docker, GitHub Actions, Vercel

education.txt

B.E. AI & Data Science

Terna Engineering College
University of Mumbai

2023 – 2027 (EXPECTED)

routine.txt / one day at a time

A scheduled illustration of my day, following India Standard Time. You can preview any moment without changing the real clock.

00:00 – 06:30
Laptop time. Build, test, repeat.
06:30 – 13:00
Sleep. Even pixels need a break.
13:00 – 14:00
Wake up & lunch.
14:00 – 17:00
Work, reading, and following a question.
17:00 – 19:00
Unwind. A game, some music.
19:00 – 20:00
Pack the bag. Get ready for MMA.
20:00 – 22:00
MMA training.
22:00 – 00:00
Chill after training.