Let's work together!

Fine-tunning open-source AI models & Agents for fun and profit

Open-source AI tiny models? Self-hosted services? What could go wrong? 😈

November 21, 2024 - 2 min. read

image-generated-using-le-chat-from-mistral
Image generated using Le Chat from Mistral

I'm launching a new experiment called NeuralNinja.com.br, tied to two simple premises:


  1. The site should be an AI training platform in Portuguese, offering just 3 things: curated AI tools, curated trainings and courses (free & paid) and news.
  2. The whole project should be created using open-source AI models, from code, to design, to images, to the content itself.

I plan to fine-tune some tiny models for different tasks, like a content generating assistant specialized in uruguayan & brazilian wine reviews to consume via a Wordpress plugin at BRAAAY), helping me out writting some Nodejs & Svelte code, generating a joker bot for Nostr, etc.

As for NeuralNinja itself, I will need to create an AI Agent to do the following:

All on a budget, on tiny models, self-hosted locally.

First step: The API

I will expose some "fine-tuned" Ollama open-source AI models using Cloudflare tunnels, exposing them via a Dockerized reverse proxy server, that I will consume myself for the content gen side of things across several projects.

I currently have just the basic setup ready, running in a local machine exposed via a Cloudflare tunnel (if the projects ends up generating any financial result I will have to setup a HomeLab for the task).

This article is a work in progress, aimed to record the experiment. One that will (hopefully) end up unleashing a tiny new AI monster into the world, and letting me learn some new cool stuff.

Next steps

ai docker ollama experiment

← Back to archive