Amar Kushwaha

I’m a computer-science student at Texas Tech who likes building and researching with machine learning and AI, and pointing it at real problems instead of toy ones. Most of what I make is software you can trust to behave — a secret scanner that blocks leaked keys before they ship, AI agents that get slashed when they break their word, a payment that clears only for work it can prove. I recently wrapped a research stint modeling how seizures spread through the brain, where I got the simulation running about a thousand times faster.

Off the clock: globally ranked in Mobile Legends, and 500+ manhwa and light novels deep. Same stubbornness, different ladder.

Now June 2026 — Applied Scientist Intern at Tempora Labs this summer, working on agentic AI.

Selected work

git0x on npm · 70 detection patterns

Blocks leaked API keys and secrets before they ever reach a commit.

npm source
Kayden ETHDenver 2026 — dual-track winner

A permanent on-chain home for an AI agent — its own identity, wallet, and a self-running heartbeat.

demo source
Sonsensus 21 verified payments on Base Sepolia

Pays an AI worker by the checkpoint — USDC streamed in proportion to work it can prove.

source contract
NeuroFormer research-grade Python library

Reads psychiatric signals from raw EEG with a hybrid graph-and-transformer model.

source
Universal Sentinel GDACS · NASA · NOAA

An AI-gated treasury that releases disaster aid from live feeds — and defaults to no.

demo source

Also building

Oath — a stake-backed promise an agent signs before it acts; break it, get slashed on-chain · try it →

ZeroGate — a one-payment paywall verified by an in-browser Polkadot light client, no accounts

Kartein — a spending firewall that checks every agent payment against your rules before it settles

Sorting Barrier — Dijkstra vs the 2025 algorithm that finally beat it, raced on a live map

full archive →

Research

Epileptic brain networks Texas Tech · Honors research

Simulating how a seizure spreads across the brain — ~1000× faster after a Numba rewrite, up to 998 regions.

source
RAWRB research benchmark

Does a small model’s step-by-step reasoning actually drive its answer, or just rationalize it?

paper try it →

Writing

delete the oracle 2026 · 06 i built a trusted oracle to catch my agent misbehaving, then deleted it and the system got safer. worker for throughput, main thread for latency 2026 · 06 “always run heavy work off the main thread” is good advice that is sometimes exactly wrong. every brain is different 2026 · 06 the fix for “the test subjects are different humans” wasn’t a bigger model — a whitening trick from 2019 and a method from the 2000s.

all writing →

Recognition

Education

B.S. Computer Science, minor in Mathematics — Texas Tech, expected December 2027. Honors College, 4.0 GPA.

Contact


github.com/kushwahaamar-dev
x.com/amarkushwaha__
linkedin.com/in/kushwaha-amar
résumé (pdf)