I’m excited to share something new. I’ve been experimenting with fine-tuning large language models (LLMs) for domain-specific knowledge, and the first public demo is now online:
This tool allows you to ask questions about topics, guests, and discussions that took place on That UFO Podcast between 2021 and 2025. The model has been tuned using the episode transcripts, so answers are grounded in what was actually said on the show.
Why I built this
For years I’ve been fascinated by how LLMs can move beyond general answers and start providing insights within a narrow field. UFOs and UAPs are a perfect test case; there’s a lot of discussion, speculation, and evolving knowledge, but it’s easy to lose track of what was actually said and when.
With this expert, you can quickly check when a topic came up, what a guest mentioned, or how certain themes evolved over time.
Under the hood
This project is 100% open source powered. Some of the tools that made it possible:
- curl → the humble command-line workhorse making requests and testing APIs
- Whisper → open-source speech-to-text model for turning podcast audio into usable transcripts
- OpenWebUI → a flexible web interface for running and serving local models
- Ollama → model management and deployment made simple
- PHP, JavaScript, and other glue code tying everything together
The aim is not just to provide a working demo, but also to showcase how open tools can be combined to produce something new and useful.
What’s next
This is very much a work-in-progress. Hardware is limited, so replies may be a little slow and accuracy isn’t always guaranteed. But if the idea proves useful, I’d love to extend the approach:
- More podcasts or shows
- Additional UFO/UAP resources
- Other niche domains where focused knowledge would be helpful
Try it out
You can explore the demo here:
👉 https://www.nordic-ufo.org/ThatUFOPodcast/
If you find it helpful, let me know! Feedback, ideas, and even small donations to help keep the servers running are all welcome.
