Open to internships

I make AI features
that work
in real products.

I'm Shubh, a third-year undergrad at IIT Bombay. I work on software engineering and applied AI. I recently interned at an AI startup, shipping search, study tools, and diagram features, and now I co-lead IITB's AI/ML team.

Email:
0%

cost reduction by making the AI write code instead of drawing pictures

0%

accuracy pulling diagrams out of exam papers

0k

sample exam pages generated to train the detector

0

ways of ranking questions tested, simple round-robin won

◉ Featured project, live

Study assistant that turns playlists into quizzes

A learning assistant on Azure that turns videos and playlists into something you can study from. Ask questions, get notes, flashcards, quizzes. If one AI provider goes down it switches to another, repeat questions stay fast with cached answers, and one bad video can't sink the playlist.

Now
AI/ML Team Lead

Co-leading the team: competitions, research, builds

Sports Council

Web Convenor, institute sports website

How I build

Longer prompts never fixed hallucinations for me. What worked was moving the fragile parts, drawing, sorting, math, out of the model into small programs with checks around them. I keep the model on words and put everything that has to be exactly right into code. I run things locally once the API bill justifies it, and I write benchmarks aimed at the places most likely to break.

// Deterministic code at the center, models at the edge.

Education
IIT Bombay

B.Tech Chemical Engineering
Minor, CMInDS

8.72/ 10 CPI
01.

Experience

AI/ML TEAM LEAD · IIT BOMBAY

LEADING THE CAMPUS AI/ML TEAM

LEADERSHIP // 1 OF 3 LEADS
Active2026 — PRESENT

I co-lead the institute AI/ML team with two others: competitions, research, builds. Right now we are hiring and mentoring the next nine members.

I set the technical direction. Shipping beats studying.

RESEARCH & ENGINEERING INTERN · AUTONISE

MAKING AI OUTPUTS CHEAP AND TRUSTWORTHY

APPLIED ML // BENGALURU
MAY 2026 — JUN 2026

I worked across the whole production stack: diagram generation, answer checking, question ranking, and keeping the whole thing free to run.

[01]

MAKING AI DRAW DIAGRAMS THAT ACTUALLY WORK

MAKING AI DRAW DIAGRAMS THAT ACTUALLY WORK
Ask an LLM to draw an SVG and the geometry falls apart. It has no sense of space, no coordinates, just plausible-looking tags. I found that out the slow way, by testing it over and...
[02]

WHY RANKING NEEDED BORING ROUND-ROBIN

WHY RANKING NEEDED BORING ROUND-ROBIN
Ranking questions by difficulty needs pairwise comparisons, and comparing everything to everything is O(N²). I tried to be clever about it: merge-sort and ELO should get that to O(...
[03]

TURNING EXAM PAPERS INTO STRUCTURED QUESTIONS

TURNING EXAM PAPERS INTO STRUCTURED QUESTIONS
I built the pipeline that eats exam papers and spits out structured questions. DocLayout-YOLO finds the text and diagram regions first. A person checks the boxes in a drag-to-adju...

Code handles space and logic. The model handles words, and the bill stays at zero.

ML RESEARCH INTERN · INFO-ME

A RECOMMENDER BUILT ON THIN BEHAVIOR DATA

RECOMMENDATION SYSTEMS
JAN 2026 — FEB 2026

A two-stage recommender built on thin behavior data. Including the part where my first evaluation metric turned out to be wrong.

[01]

RECOMMENDING FROM READING TIME, NOT CLICKS

When you barely know what a user reads, you use what you have: how long they linger. Dwell time becomes a weight, the weights become an interest vector, and cosine search pulls the...
WEB CONVENOR · INSTITUTE SPORTS COUNCIL

INSTITUTE SPORTS WEBSITE

IIT BOMBAY // STUDENT COUNCIL
Active2025 — PRESENT

I run the institute sports council website: events, schedules, and contacts in one place.

[01]

WEBSITE DEVELOPMENT

WEBSITE DEVELOPMENT

I built it responsive. Students check it for fixtures, tournament schedules, and who to contact.

[02]

PORTAL MAINTENANCE

I keep the content fresh and the portal running.

02.

Skills & Tools

ML SYSTEMS

LLM Systems & PipelinesProgrammatic IR (LLM → Code)Evaluation MethodologyInference OptimizationRanking Systems (Swiss-Tournament)Recommendation Systems

ENGINEERING

Pipeline ArchitectureHybrid Local/Cloud InferenceEmbedding-based RetrievalAPI IntegrationsDeployment (Vercel, Railway)Frontend (React / Next.js)

STACK

PythonMatplotlib · RDKitvLLM · LoRA · MoETypeScript (Working Knowledge)Linux

MINDSET

Technical & Team LeadershipSystems ThinkingFailure Mode AnalysisMetric Design & CritiqueBenchmarking & ExperimentationResearch ↔ Engineering Bridge