karmedge

Robert Lukoszko · 23 · @karmedge

Building an LLM OS your business runs on.

The way companies run on spreadsheets and virtual machines today, they'll run on an LLM OS tomorrow — the layer where the model, your data, and the interface meet. I'm building it, with a focus on generative UIs: screens that compose themselves, in real time, around the task at hand.

Get in touch Follow on X GitHub

About

The bottleneck is the interface.

Most software fixes the shape of an interaction before you even arrive — rigid forms, rigid tables, rigid buttons. Generative UIs give that up: interfaces that render themselves from the conversation, the data, and the model. That's the foundation of an LLM OS. A business shouldn't have to choose between a locked-in SaaS interface and writing code. It should just be able to ask.

Generative UI
Interfaces that compose themselves at runtime.
Agents
The model doesn't just answer, it acts.
Vector search & embeddings
Client-side, WebGPU, llama.cpp. Fast and private.

Work

Things I've shipped

dosearch.me
The company behind the LLM OS work in progress.
BakLLaVA Realtime
Stream your webcam; a vision-LLM narrates it live. Went viral on X.
Chat with Page
Chrome extension. Your OpenAI key, your data. 100% private.
Client Vector Search
Embeddings in the browser. No server calls.
Stormy AI
YC-backed. End-to-end AI influencer outreach, automated.

Videos

Talks & demos

My AI Tinkerers / YC lecture, plus where I demo.

From Viral AI Demos to YC — Robert Lukoszko 55 min · AI Tinkerers One-Shot
Featured talk
From Viral AI Demos to YC
Three days before GPT-4 Vision, my BakLLaVA webcam demo went viral. This is the full story: local models, distribution, and pivoting to a venture-scale business.
AI Tinkerers · One-Shot writeup
The long-form companion to the talk.
AI Tinkerers SF
The community I demo at.

Listen

While you read

Coding is a contemplative act. This is on loop.

Contact

Building a generative UI or an LLM-native product? Let's talk.

Email me Telegram X LinkedIn