CSE UNDERGRAD · MIT MANIPAL

I like to figure
things out.

I’m Kaustubh. I build with code, experiment with AI, and follow a good question wherever it leads: from algorithms to products to real-world problems.

Explore selected work Or say hello

BASED IN INDIAALWAYS LEARNING, OFTEN BUILDING

IDEA → EXPERIMENTFIG. 001

AThink in systems

BLearn by making

OBSERVE / BUILD / REPEAT↗

01INTRODUCTION

01 / A NOTE ON HOW I THINKTHE WHY BEFORE THE HOW

Curiosity is the
starting point.
Building is the proof.

I’m studying Computer Science and Engineering at MIT Manipal. I like understanding how things work, testing ideas instead of just talking about them, and finding the useful connection between a technical possibility and a real problem.

That might mean training an agent to play a game, designing a smarter way to solve Wordle, or exploring an early-stage idea around recycling. I’m also curious about the products, markets, and decisions that help good ideas find their place in the world.

A little more on LinkedIn

02 / SELECTED WORKTHINGS I’VE BEEN FIGURING OUT

Ideas, in progress.

Different questions, different tools. Each project is a chance to learn something by making it real.

01 — DECISION MAKINGLIVE EXPERIMENT ↗

SALET SHARE CRANE

INFORMATION THEORYENTROPY / MINIMAX

AI & ALGORITHMSPROJECT 01

Wordle, treated
as a decision problem.

A solver that chooses guesses strategically instead of relying on instinct.

THE QUESTION

Which guess gives the most useful information about the hidden word?

THE APPROACH

Compare entropy-based information gain with a minimax strategy to narrow the candidate space.

PythonInformation theorySearch strategies

Try the live solver ↗View source ↗

02 — LEARNING BY PLAYINGAGENT STUDY

2048 / DQN

24248162326448128256

STATEACTIONREWARDREPEAT

REINFORCEMENT LEARNINGEXPERIMENT / EVALUATE

AI & REINFORCEMENT LEARNINGPROJECT 02

Teaching an agent
to play 2048.

An experiment in getting a game-playing agent to improve through experience rather than a fixed list of moves.

THE QUESTION

Can an agent learn useful board-management decisions from repeated play?

THE APPROACH

Build a DQN-based agent and evaluate learned behavior against the game’s changing states and rewards.

PythonDeep Q-learningModel evaluation

Live Demo ↗View source ↗

03 — TECHNOLOGY WITH PURPOSEEARLY STAGE

BOTTLE RECOGNITION / CONCEPT

R
↻

AI / COMPUTER VISIONRECOVER WITH TRACEABILITY

EARLY-STAGE STARTUP INITIATIVEPROJECT 03

rebottled: exploring a smarter loop.

An early-stage sustainability initiative exploring how AI-powered reverse vending could make bottle recovery more traceable.

THE OPPORTUNITY

Make it easier to identify returned bottles and create useful recovery records.

THE EXPLORATION

Computer vision for bottle recognition, recovery logging, and potential support for recycling and EPR workflows.

Computer visionProduct explorationSustainability

Ask me about rebottled ↗Early-stage initiative; details evolving

NOTE / These projects are works in progress. Descriptions focus on the problems and approaches explored, not unverified performance claims.

03 / MY WORKING LOOPFROM QUESTION TO SOMETHING REAL

Make a guess.
Test the guess.

I like projects where the next step isn't obvious. A simple loop helps me turn curiosity into progress.

Ask better questions

Understand the problem before jumping to a solution.

↗

Build a small test

Turn an idea into something concrete enough to learn from.

⌁

Observe what happens

Look at behavior, trade-offs, and what the experiment actually says.

↻

Iterate with intent

Keep what works, question what doesn't, and try again.

04 / TOOLS & CURIOSITIESWHAT I’VE BEEN WORKING WITH

Technical foundation.
Open tabs everywhere.

I learn best by building. These are tools and areas that have appeared in my projects and experiments, rather than a rating of my proficiency.

LANGUAGES & DEVELOPMENT

PythonCJavaGit / GitHubHTML / CSS / JS

AI & COMPUTING

Reinforcement learningComputer visionAlgorithmsInformation theoryModel evaluation

ALSO CURIOUS ABOUT

Product strategyStartupsMarket researchConsultingHuman-centered techCookingMovies & TV

05 / THE JOURNEY SO FARSTILL EARLY. STILL EXPLORING.

Learning in
public.

A few threads that connect what I study, what I build, and what I want to understand next.

FOUNDATION

Computer Science & Engineering

Studying at MIT Manipal and building foundations in programming, systems, and problem-solving.

BUILDING

AI and algorithmic experiments

Exploring Wordle strategy, a DQN-based 2048 agent, and personal assistant tooling through hands-on projects.

EXPLORING PRODUCTS

Entrepreneurship Cell

Contributing to startup research, event publicity, and student entrepreneurship activities.

NEXT QUESTIONS

Technology, products, and strategy

Learning how technical ideas become useful products, and how teams make good decisions under uncertainty.

06 / OPEN CHANNELQUESTIONS, IDEAS, COLLABORATIONS

Got a good
question? So do I.

If you’re building something interesting, exploring an idea, or want to talk AI, software, products, or problem-solving, I’d be happy to connect.

Email me ↗WhatsApp ↗LinkedIn ↗GitHub ↗

EMAILkaustubhs265@gmail.com

PHONE / WHATSAPP+91 7275472411

I like to figurethings out.

Curiosity is thestarting point.Building is the proof.