Student · builder · researcher

I build tools that
measure the real world
and get used.

Hours logs, instrument libraries, incident maps, spacetime, scholarships, model drift — most of what I make exists to track something and put it in front of the people who need it. A lot of it is running in schools right now.

450+students using my tools
7schools
🥇🥈BU & Microsoft hackathons
2startups I'm building

Startups

Two things I'm building into companies.

Tools people use

Software running in real schools and competitions. Each has a click-through test site with fake data — no login needed.

Runs in your browser

No screenshots — the real thing, loaded on demand.

Also built

A desktop simulator and a theme that traveled further than expected.

Hardware & systems

Things I've built with my hands and my own hardware, going back to age seven. Photos and write-ups are on the way.

Hackathons

Weekend builds that placed — two of them, plus the one that became a company.

Leadership

Where I've learned to run things and bring people along.

Milestones & events

A few that matter — with more detail to come.

Research

A paper I'm writing.

paper in progress

Controlling Parameter Drift in LoRA Continual Test-Time Adaptation via Reset Detection

When a model adapts to new data at test time, it can drift — over-fitting to a shifting domain until it forgets what it knew. I'm studying whether freezing the base weights and adapting only the LoRA layers keeps that in check, and whether watching the model's prediction entropy can detect the moment drift sets in.

When entropy spikes past a threshold, the system dynamically resets the adaptation. The question: does reset detection help more as the domain shifts get more severe? That's what the experiments are built to measure.

LoRAContinual test-time adaptation Entropy-based drift detectionDynamic reset
test stream → model base — frozen LoRA — adapts entropy monitor drift → reset
Base weights stay frozen; only LoRA adapts. When the entropy monitor sees drift, it resets.

About

I'm Neel Bansal. Most of my projects start the same way — I notice a thing people are doing badly by hand (texting one person to log 150 students' service hours; losing instruments to damage nobody reported; scattering sheet music across a dozen Google Classrooms) and build the system that should have existed.

Some of it is competition work: a scholarship-matching platform that took 2nd at a Microsoft hackathon, and a public-safety awareness map that won 1st at Boston University. Some of it is just curiosity — a relativity simulator, a programming language made of slang, a paper on how models forget. And two of them — Freshline and Instrumentify — I'm turning into companies.