ak

📍 Pilani, India

ON MY DESK RIGHT NOW

A tiny ML interpretability library

for our prof's research group.

Interpretability for small language models.

Working with Dr. Priya Sundaram (CS dept., BITS Goa) on visualizing which neurons activate for which concepts in 1B-param models. I write the experiment code; she keeps me from drawing the wrong conclusions.

A CLI that summarizes long PDFs locally.

Open-source tool that runs a quantized Llama locally and gives you a one-page summary of a 100-page paper. I built it because the cloud version cost me ₹400 a week. Now ~80 stars and counting.

Distributed systems, slowly and properly.

Reading DDIA cover-to-cover this semester. Implementing each chapter's ideas in a small toy. Currently on Chapter 7 — distributed transactions, two-phase commit, and the long history of getting them wrong.

Studybuddy

Find a study partner on campus before the exam, not after.

A small web app that matches students by course code, hostel block, and preferred study time. Built after watching three friends fail the same midterm because they thought everyone else was "too good to study with." Currently 280 weekly active users at BITS Goa.

Glassbox

A library for visualizing what tiny transformers are doing inside.

Built as part of Dr. Sundaram's research group. Wraps a 124M-param GPT-2 and surfaces per-token attention heatmaps in your terminal. We're using it internally to debug interpretability experiments before sending any work to a paper.

Foundwell

A health-camp dispatcher we built for rural Rajasthan.

Smart India Hackathon 2024. A team of 4 — we won 2nd place out of 380 teams. Maps PHC capacity to incoming patient load using a small ILP, sends SMS triage instructions in Hindi. We slept four hours total. Built fast, slept ugly, surprisingly proud of it.

xv6-extensions

A semester of pretending to be an OS engineer.

For Prof. Rajiv Mishra's Operating Systems class. Added copy-on-write fork, a slab allocator, and a tiny scheduler tracer to the xv6 teaching kernel. Got the highest marks in the class on the final project, which mostly means I spent more weekends debugging page faults than I'd recommend.

B.E. (Hons.) Computer Science

BITS Pilani, K. K. Birla Goa Campus

CGPA

9.21 / 10

Top 5% of class, last 3 semesters

RESEARCH

Aug '24 — present

Undergraduate Research Assistant

Interpretability Lab · Dr. Priya Sundaram (CS dept., BITS Goa)

Approached Dr. Sundaram after taking her Deep Learning class — she said yes after I sent her a small open-source repo I'd been hacking on. I now run weekend experiments for the group, maintain our Glassbox toolkit, and am co-authoring a workshop paper for Q1 2026 on circuit discovery in 1B-parameter models.

SUMMER 2025

10 weeks

Bengaluru · remote-friendly

A Series A developer-tooling startup. I was the only intern on the platform team. Shipped a CLI rewrite (cut cold-start latency from ~1.2s to ~280ms), added retry-with-backoff to the SDK, and wrote the first version of their integration test harness. Two of my PRs are still running unchanged in production.

Deep Learning Specialization

Andrew Ng · DeepLearning.AI

2024

CS50 — Intro to CS

David Malan · Harvard / edX

2023 · Honors

Cloud Practitioner

AWS · CLF-C02

2025 · 892 / 1000

Practical Deep Learning

Jeremy Howard · fast.ai

2024

₹75K prize

Renewed yearly

Remote-final

Ongoing

♞

Chess

FIDE 1842 · Captain, BITS Goa chess team

Started during lockdown, took it too seriously, now I play one rated game a week and analyze it on Sunday morning. Best opening: King's Indian. Worst flaw: I trade pieces when I'm tired.

✿

Reading

Mostly science fiction. Currently: A Memory Called Empire.

I read about 30 books a year, almost all SF and history of science. The best book I've read recently was Stuart Russell's Human Compatible. Ask me about it — I will not stop talking.

✦

Running, badly

10K PB: 51:14. Aiming for sub-50 this winter.

I started running because I needed something I was unambiguously bad at. Three years later I'm still bad — but I keep showing up at 6 a.m., which I'm starting to think is more useful than being naturally talented at things.