Vibhum Aggarwal
student at BITS Goa. i build things.
i'm vibhum. i build AI products and do research — right now that's hunch, an AI that texts you first, and computer vision at a lab at BITS.
most of it comes down to one question i can't shake: wait, can you actually do that? — usually asked around 2am. some of the answers ship, some stay research, and a few just teach me why they don't work.
- founder-in-residence at Forge
- consultant at 180 Degrees Consulting, BITS Goa
- building hunch, an AI that texts you first
- tracking reactor bubbles with SAM2 at a BITS research lab
- back at BITS Goa after a summer at Berkeley
updated october 2026
hunch — a friend that lives in your texts2026
a personal AI agent that lives in your iMessage, self-hosted and free to run. you text it like a friend and — the part i care about — it texts you first. reminders, follow-ups, the 9am "hey, that landlord email?" nobody asked it to send.
crewline — a small team of AI workers on your own repos2026
a local command center for persistent Claude Code workers. each one has a role, explicit permissions, its own git worktree and memory that survives between sessions. you assign tasks from one box, approve the risky bits — running tests, committing — from inline cards, and workers hand follow-up work to each other. everything stays on your machine.
download & run it →code ↗or click the name to see it running
raahai — a voice-first guide for getting around2026
you talk to it, and it plans your commute and errands from live ambient data, then narrates the world as you move through it — a low-latency companion for solo travel and the daily grind of getting places.
click the name for the pitch
nazrai — airbnb for billboards2025
an AI for out-of-home advertising. owners list their billboard; a client gives a budget, a timeline, and who they want to reach; and nazr works out the actual OOH strategy — end to end, no media buyer required.
click the name for the pitch
reelvault — files, hidden inside video2025
store any file or folder inside a video and get it back byte-for-byte. robust mode survives re-encoding, uploads, even a screen recording. runs locally in your browser tab.
youtube-sentiment — what a comment section feels2025
scrapes a video's comments and reads their sentiment — VADER scoring plus embedding clustering to surface the real opinion groups and where the mood shifts.
catnap — ML on antibody neutralization2026
predicting whether an antibody neutralizes HIV from sequence alone. the honest finding: our early strong numbers came from antibody leakage between train and test, and a fine-tuning "collapse" turned out to be an fp16 overflow bug, not the learning rate. split by antibody and most of the magic disappears. the real work is being the one who checks.
reactor bubble tracking — SAM2 + YOLO2025–now
with PhD scholars at BITS, segmenting and tracking bubbles flowing out of a reactor — teaching a visual foundation model to follow something that's constantly merging, splitting, and disappearing.
physics-informed nets — university of delhi2024–25
with Dr. Sangeeta Yadav: an LSTM air-quality predictor under 2% mean absolute error, with synthetic data to repair corrupted datasets — then a CNN–PINN fusion modeling blood flow and pressure from CT scans, aimed at detecting aneurysms.
- why hunch is picky about modelsoct 2026
- ten days of silence2024
- a brick, duct tape, and a championship2024
- my gap year, mostly2024
i'd been fascinated by colleges abroad since i was a kid. last year i got into a few and didn't go — i chose BITS, partly for my own reasons and partly for ones that weren't mine to pick. that sat with me for a while.
so when my freshman year ended, i went to the US anyway. classes at UC berkeley in the morning, san francisco in the afternoon. two months of luma events, free dinners, people much further along than me. the thing that actually changed something was them — everyone i met was high-agency, never stopped. i came back wanting to build a company, and a few months later i did.
read the whole thing →collapse ↑
i enrolled in summer sessions at UC berkeley — game theory and data science — which was, in a roundabout way, the thing i'd wanted since i was a kid. i wanted new york too, which meant two visas and more paperwork than i'd thought about. i applied without really stopping to think, and it came through in time.
the classes were the reason i was there and the shape of every morning. the rest of the day i had no plan for at all — which turned out to be the important part. berkeley until the afternoon, san francisco after.
what followed was two months of events, free food, and people much further along than me. i lived on luma — scrolling it between lectures for whatever was on that evening, going for the free dinner and staying for the conversation. founders, VCs, rooms i had no business being in. i went flat broke anyway, which felt like a fair trade.
i also flew halfway around the world and ended up at a diljit dosanjh concert, which is funnier the longer you think about it.
i got out of the city too — stanford, tahoe, and seattle over the 4th of july, where some of my school friends had ended up. seeing them again, a year on, in that setting. somewhere in all of it i met suhas sumukh, and a lot of people i'm still messaging now.
i sat in a waymo for the first time and watched the steering wheel turn by itself, which is the closest i have come to living in science fiction.
i didn't own a laptop until covid. before that, almost everything i learned, i learned by sneaking into a room i wasn't allowed in.
when i was 7 or 8 my school had a robotics lab — but it was only for the middle and high schoolers. i was in junior school, so i technically wasn't allowed near it. so i'd sneak in during lunch breaks and just watch, and slowly started picking it up. in grade 4 i built my first real thing: a bluetooth remote-control car. it took four months, because the only time i could work on it was lunch, in that lab, with my robotics coach. after that came every arduino project i could think of — an ultrasonic radar, smart-home systems, IoT gadgets, drones — including one meant for disaster relief.
mostly, though, i was the kid who had to open everything. in grade 5 i took my camera apart, found it had rusted up inside, cleaned it out — and it just worked again. that was weirdly enlightening. after that it was the home theatre system, then DC motors, which i'd crack open to study the windings and rewind them. not everything survived my curiosity: once my dad ordered LED lights for the car, and i got so into it i bought a whole pile of my own gear too (sorry dad, it wasn't cheap), then plugged the LEDs straight into AC power just to see what would happen. it went about as badly as it sounds.
that same year i got into robo wars, and i never really stopped. from then on it was combat bot after combat bot — building them, getting them wrecked, rebuilding them, and hauling them to competition after competition. my first win came in grade 7, and i was so hyped i went straight to a national competition at IIT Delhi, where i got knocked out in the first ten seconds. somehow that's still one of the moments that shaped me most.
i kept at it all through school and ended up vice president of RoboKnights, our robotics club. one of my favourite stories from those years: i was sent to help a novice at an inter-school competition, and their robot was a wobbling, lopsided mess, barely built to stand. with seconds to spare i patched it with what was lying around — a brick and a lot of duct tape. it rolled out against sleek, polished bots, and the whole room laughed. until we won. (i wrote about it.)
i didn't have a laptop, but i had youtube. i watched everyone i wanted to learn from — Kunal Shah and Nikhil Kamath since 2019, before they were everywhere, and Elon, because my dad pointed him out to me when i was 7 or 8, back when he was nowhere near the richest person in the world.
then covid hit, i finally got a laptop, and the whole internet opened up at once. i went a little feral with it. i started a premium eyewear shop — specscafe — because i figured nobody in India was selling nice eyewear online; built the catalog, built the site, and it never really took off. the sneaker side did: i built a discord cop bot, onlydrops, that members used to buy around ₹60 lakh of sneakers in its first year and ₹90 lakh in its second. i made my first internet money on Dogecoin and GameStop, then lost all of it inside a week of trying to trade crypto. i built a site that generated NFTs for you to sell. it was a lot, all at once, and i loved every bit of it.
after grade 12 i took a gap year — partly personal reasons, mostly to figure out what i actually wanted. i joined an AI research lab at the University of Delhi and spent saturdays handing out 400+ meals with the Mayur Vihar Tamil Association. i also fell off a first-floor windowsill taking a shortcut to my room — my buffalo moment. my old school invited me back to judge robo soccer at Exun 2024 — a RoboKnights event, at a tech symposium with 25+ schools. the kid who used to sneak into the robotics lab was now the one judging. i applied abroad, got into some good schools with scholarships, and chose BITS Goa anyway.
BITS overwhelmed me. for the first time i had no idea what i wanted, so i did everything — joined every society and club i could. three or four months in, it clicked: i needed to build. i found my co-founders and we started raah AI; it began as a token predictor for billing AI-model subscriptions, then one idea turned into the next, and the next. that path led me to Forge as a founder-in-residence, surrounded by people genuinely killing it — and that's where entrepreneurship really clicked.
what's on my shelf, in my ears, and on my screen.
books
- Factfulness ★
- How to Be a Bawse ★
- The Almanack of Naval Ravikant
- a stack of Star Wars & Marvel paperbacks
+ 7 moreshow fewer
- Measure What Matters
- The Intelligent Investor
- The Power of Your Subconscious Mind
- The God Delusion
- Facebook: The Inside Story
- Quantum Field Theory for the Gifted Amateur
- The Little Book of Data
watching
- PewDiePie
- MKBHD
- Kelly Wakasa
- random podcasts
listening
- Coldplay
- Taylor Swift
- When Chai Met Toast
- The Local Train
reading online
this bit's for people who actually know me. got the password?