06/12/2025 –, Main Stream Idioma: English
As a language teacher turned AI developer, I set out to create something that combines both of my passions: a self-hosted Swedish tutor powered by open-source models. This DIY project combines Ollama, LangGraph, and Gradio into a modular system that runs locally, delivers nuanced feedback, and costs nothing to run. No cloud inference, no API keys, just a playful, powerful way to decentralize AI and make language learning fun again.
At B2/C1 level, language learning resources tend to dry up, and Swedish is no exception. So in a true DIY spirit, I decided to build my own AI-powered tutor.
In this talk, I’ll show how I combined my past life as an English teacher with my current role in AI software development to create a self-hosted, modular Swedish learning tool. You'll get a behind-the-scenes look at why I love Ollama and how I used it to run local models, LangGraph to build agentic AI workflows, and Gradio for a lightweight, interactive frontend.
This talk isn’t just about language learning, it’s a case study in creative constraint, smart model selection, prompt engineering, and building usable AI systems with zero commercial cloud dependencies. If you're curious about multilingual LLMs, agentic reasoning, or sustainable AI development, this is for you.
Milena Miernik is a former language teacher turned AI software developer, blending her background in linguistics with a passion for building smart, human-centered tools. Based in Sweden, she holds a Master’s in sociolinguistics and now works at Aixia AB, where she develops AI solutions with a focus on language and education.
With experience teaching advanced English and learning Swedish herself, Milena brings a unique perspective to multilingual AI. She’s especially interested in self-hosted, open-source models and agentic workflows that promote transparency and decentralization. Her DIY spirit and cross-disciplinary mindset drive projects that are both technically inventive and deeply user-focused.