"You will soon embark on a lucrative business venture."

Yesterday I went to the mall down by [REDACTED] and found a fortune on the ground that said I would embark on a lucrative business venture. As someone consumed by the ideal of fate, especially when hopeless and jobless, I found myself encouraged.

I have many ongoing projects. Here is a list:

1. Epsilon

An ongoing research model I've been developing in my spare time. It explores alternative transformer structures with an emphasis on interpretability, sparsity, and efficient computation. While I'm still working on fine-tuning, it incorporates several non-standard design choices inspired by prior work on dynamic computation and symbolic processing. So I'm currently working on documenting and understanding it more rigorously before considering any form of public release. I plan to revisit key fundamentals in machine learning to ground myself more deeply in the theory. But if you're curious, you can look at this research paper that I've read that somewhat helped me brainstormed a part of it: Adaptive Computation Time (Graves, 2017).

2. Pingaroo and its API

Pingaroo or its API counterpart that I ripped out of it (AI-powered JSON creator) is one of my stronger SaaS products. It's essentially an AI-assistant for your email inbox for people who don't want to miss an important/urgent email. Likely, there will be a free tier, $10 standard tier, and a $20 pro tier. More details on its page.

3. VisionTag

One of my more lightweight and more viral-prone ideas that was inspired by the latest controversy. You can use a chrome extension to right-click any image and have an option to "Scan with VisionTag" to check if it's AI-generated. However, there are current roadblocks such as; being unable to scan YouTube thumbnails because hovering over a video creates a preview (becomes janky, Gemini was unable to come up with a creative idea) and people tend to expect simple tools such as these to be free. However, as much as I would like this to be mostly free instead of three free image detections before you upgrade, the API I use from SightEngine costs money. Also I was thinking I could potentially create a YouTube video AI detection via API takes sparse screenshots throughout video, while cost ineffective, could potentially encourage people to look into it and tip me~

4. CodeTag (Codebase Compression)

A spontaneous project. Imagine if you could compress a 1M token codebase directory folder into 250k tokens or so, and get a json filled with the codebase's structure. Very useful for LLM workflows, I'm looking at you Gemini. I'll be open sourcing this project on GitHub.