Summary

I build self-hosted AI applications using React.js, Node.js and WebRTC, specializing in connecting Ollama and other local models for collaborative environments. My technical stack includes experience with OpenAI API, Hugging Face frameworks (Smolagents, LlamaIndex, LlamaGraph), Pydantic AI, and Python for streaming model outputs directly to browsers.

Details

My main focus is on self-hosted models, both LLMs and image generation, as well as distributed networking, such as WebRTC. The projects I’m most interested in are ones which empower individuals or small groups to run their own models.

I have also worked with the mainstream models’ APIs and the tools commonly used with them. I’m currently learning three agentic frameworks through the Hugging Face course (Smolagents, LlamaIndex, and LlamaGraph) and learning Pydantic AI on my own.

I’ve tested multiple TTS systems locally and written about this on my blog.

I placed third in the Apart Labs AI hackathon last year, and was invited to participate as an Apart Labs fellow. Our team built a benchmark to assess model’s ability to geolocate based on images.

I’ve built two browser-based platforms which give users the ability to have a live group AI session, connected via WebRTC. The first (HaveWords) used the OpenAI API, and focused on group roleplay and creative fiction. The second (AgentCircus) connects users together and allows them to connect their own models to the session using Ollama, and allows users to chain model responses among the different participants. These are built with React.js and Node.js primarily.

I’m also building a python project (StrapTo) which allows users to stream the outputs of self-hosted models over WebRTC directly to browsers.

I’m less focused on the particular role that I might find, and more focused on finding a good team of people working on something interesting. I’d like to hear what you’re working on!