I set out to see if a model could learn something new without retraining. The result is a working prototype that uses virtual tokens and LoRA adapters to hot-swap micro-models in real time through a Streamlit interface.

Lately, I've been exploring an idea that sits somewhere between RAG (Retrieval-Augmented Generation) and fine-tuning, but with a twist: what if we could dynamically expand a model's “memory” by injecting data as virtual tokens instead of through traditional context windows or retraining?

When users talk to AI, they speak like humans. That’s the whole point. But the systems powering those conversations—whether it’s Azure DevOps, a database, or a custom app—speak a different language. This is where intent requests come in.

Once I wrapped my head around embeddings, I started looking into how existing products were implementing them.

I officially started working on Empirical, and things are shaping up! I'm building it with Ionic React and TypeScript

I was able to finish my "mass production" of upcycled reusable grocery bags. There are 20 bags in total. On average it took me about 1 1/2 hours per bag.