The thinking behind everything.

HackGT was the first time I decided to just go big with my hack. We built Caladrius, a privacy-first AI triage assistant for hospitals. We worked with encrypted medical passports on mobile devices,…

5 sections · 736 words

FORMulator was the first hackathon I did at Purdue, and I was super excited. The idea started as Just Dance but using a camera that tracked your pose and compared it in real time. Midway through, we…

3 sections · 243 words

I got tired of the same problem everyone hits: you want your AI agent to touch Google Workspace, and suddenly you're wiring up 4-5 separate MCP servers, each with their own auth flow, their own…

8 sections · 1,120 words

This project started in an unlikely place: a high school laboratory. Using lasers, polarizers, beamsplitters, and a handful of other optical components, we built and aligned a prototype capable of…

5 sections · 886 words

Instead of worrying about the accuracy-versus-fun tradeoff, I focused on the simplicity of the concept. For World Quantum Day, in a world where most people don't even know what quantum really means,…

7 sections · 874 words

Working on Repple has been the most fun I've had working on an app. Together with Arjun Chadha, Rishab Chakravarty, and Ayush Guhan, we shipped fast and turned the gym into a game.

12 sections · 1,129 words

In high dimensional data, distances are large, and thus, clustering can often be led astray by the sheer magnitude that Euclidean distance scales to. Self-organizing maps reduce dimensionality while…

6 sections · 363 words

We built Verbatim at BoilerMake — a video understanding platform that goes beyond simple transcription. We didn't win, but our frustration turned into determination, and we took our hackathon project…

5 sections · 345 words

Clinical trials have a serious trust problem. An estimated 4% of volunteers are "professional patients," and up to 25% of trial data may be problematic or fabricated. Current fraud detection relies…

7 sections · 743 words

It's so incredibly difficult to gather good data for drone imagery. We had to do multiple test flights to even get enough to fine-tune a YOLO model. On top of that, the real difficulty was flying the…

6 sections · 439 words

It was the model, to be honest. Trying to write a model with minimal public code out there, outdated libraries, and duct-taped GitHubs that were written in languages people don't use and set up for…

5 sections · 339 words

The SDK was honestly just a wrapper around a shaky API. It required some duct taping, but it was nothing special. It just made it easier to use in development. The work was about retry handling,…

6 sections · 390 words

This was the first time I did ML research at a job, and it was incredible seeing a problem that required deep technical ML work. The problem was reconstructing transmission loss graphs from a sound…

3 sections · 612 words

During the summer, I began work on a new kind of malware detection program. Our program was a first-of-its-kind reinforcement-learning agent to intelligently navigate complex IoT device environments.…

4 sections · 525 words

I'm building ambient AI, AI that's always on, can handle asynchronous tasks, and processes events that happen in your life without having to explicitly point it to doing something with each new item.

18 sections · 1,266 words

The main problem we aimed to fix at BuildPurdue was that there was no density of founders. What that meant was that Purdue was producing a lot of great founders, but they weren't in the same place.…

13 sections · 1,675 words

When agents save time and money, not just effort.

Agents are genuinely useful when they actually save you time or money. I absolutely do not mean save you effort. There are so many agents in the world that save people effort, which is useless,…

5 sections · 338 words

The duct tape test: when the whole point is for a human to do it.

A use case is duct tape when the entire intention of something is for a human to do the action.

2 sections · 189 words