ArnavSinghal
Making computers keep their promises: distributed systemsAI productsquantum research
software engineer · delhi, india
what landed so far
Proof of work.
Five things that actually happened. Remix the visuals all you want, these stay.
Google Summer of Code
Google paid for a summer of open quantum-ML research with ML4Sci.
National Hackathon Winner
Smart India Hackathon. First place out of thousands of teams, nationwide.
Amazon ML Summer School
Picked from a national applicant pool.
Best Paper Award
Peer-reviewed, published with Springer, best paper of the venue.
UnifyApps
Shipping product infrastructure to enterprise AI customers.
✂ snap
ONE BROKEN THREAD IN A MILLION
rewound & rewoven
SAME THREAD. SAME PLACE.
act i · the problem
One bug in a thousand runs. Gone when you look.
A big software system is thousands of threads crossing at once. Sometimes one snaps. Rerun the system and the snap hides. Nobody can fix a failure they can’t see twice.
act ii · what i built · weft
Chaos, combed into order.
Weft re-runs all that chaos in perfect, scripted order: every message, every crash, on a fixed schedule. If a thread is going to snap, it snaps on demand, right where you can watch.
act iii · the result
A real bug, caught red-handed.
Weft ran five hundred controlled storms, checking seven rules of correctness at every step, and caught a real, documented failure in a classic distributed protocol. Not a demo bug. The genuine article.
500
controlled storms
✓ true story7
rules checked, always
✓ true story1
real bug caught
✓ true storyact iii · the proof
Same start. Same snap. Every single time.
That’s the superpower: perfect replay. No “works on my machine.” No bug that hides when you look at it. Rewind the world, run it again, and the flaw shows up in exactly the same place.
- Rust
- Deterministic replay
- Fault injection
- Chord DHT
project 02 · cliniqops
CliniqOps writes the doctor's notes. So doctors can doctor.
It listens to the visit, writes the note, tags the billing codes, and files everything before the patient reaches the parking lot. Scroll to sit in on a consult.
act i · the problem · the old way
Typed by hand, hours after the visit.
Every consult ends with a doctor at a keyboard, reconstructing the conversation from memory while the next patient waits.
act ii · what i built · live in the room
Good morning, what's brought you in today?
This pounding headache. It's been three days now.
Bright light makes it worse, and I've been nauseous since yesterday.
No fever, and your blood pressure reads 118 over 76, that's healthy.
This presents as migraine without aura. Let's start sumatriptan and review in two weeks.
act iii · the result · the note writes itself
Pounding headache × 3 days
Photophobia · nausea
BP 118/76 · afebrile
Migraine without aura
Sumatriptan 50 mg PRN · follow-up 2 wks
filing to the hospital’s system
filed ✓
project 03 · q-maml · google summer of code
Q-MAML: particle physics, taught to a quantum computer.
act i · the problem
Too much lightning to look at.
CERN’s detectors record millions of particle collisions a second, far more than any human could ever look at. And finding the rare, interesting ones means retraining a model from scratch for every new question.
act ii · what i built · q-maml
Lightning, taught to aim.
Google paid for a summer of teaching a quantum computer to learn how to learn, trained on 27 terabytes of the real thing. Watch the debris: every fragment gets sorted in real time. Green is a keeper. Violet is noise.
- Quantum ML
- Meta-learning
- ML4Sci
- CMS · LHC
act iii · the result
+15%
sharper than the classical way
✓ true story−80%
less training to learn new tricks
✓ true story27TB
of real collision data from CERN
✓ true storythe climb · 2024 to now
Seven milestones, in order.
2024
DRDO
Research Intern
Applied research internship with India's Defence R&D Organisation.
2024
Springer
Best Paper Award
Published research recognized with a Best Paper Award.
2024
Smart India Hackathon
National Winner · First Place
First place, nationwide, at India's largest hackathon.
2025
Google Summer of Code
Contributor · ML4Sci
Q-MAML: quantum meta-learning for high-energy physics, validated on 27TB of CMS collision data.
2025
Amazon HackOn
Semifinalist
Semifinalist at Amazon's national HackOn hackathon.
2026
Amazon MLSS
ML Summer School
Selected for Amazon's competitive Machine Learning Summer School.
since 2026
UnifyApps
Product Engineering Intern
Building product infrastructure at an enterprise AI platform. Real customers, real stakes.
last scene. your move