October 2026
Some people weren't "getting" the different search modes in my original Armchair Expert AI Podcast Explorer.
So I integrated Jev to change up the user experience.
……….now they don't need to "get it". They just use it.
TL;DR
- Two years ago, I built an AI Podcast Explorer to teach myself how to build an AI agent using OpenAI's APIs.
- I've now integrated Jev to remove the need to choose between search modes.
- User error is now zero!
- I also upgraded the search with a vector database and semantic matching.
Give it a go here >> Armchair Expert AI Explorer
Why is this cool?........Let's start with the basics.
What is Jev?
Think of Jev as a little AI helper that interprets a request and helps an app decide what to do next.
More technically, it's TypeSafe's first System One model: built for fast, focused decisions, returning structured choices, scores and probabilities.
It works alongside an LLM or agentic application.
An LLM can already make these decisions, but using one for a simple routing task can feel like hiring an orchestra to be your doorbell.
Jev gives you a tool built for that small, specific job.
The problem: people don't read instructions…
The previous interface asked users to understand three search modes (episode, general and keyword) and switch between them during a conversation.
That added friction. Users had to understand how the tool worked before they could get the most out of it.
What Jev brought to the table
Complete simplification in one swipe.
Users ask their question. Jev helps the app choose the search workflow.
For example: "What are all the books and authors discussed with Adam Grant?"
Previously, you would need to:
- Find the episode.
- Ask the AI to list the books.
- Switch to general search to fill gaps, such as missing author names.
Now, Jev identifies which search capabilities the request needs. The app uses those decisions to run the workflow and return a combined answer, or ask a clarifying question.
What took several interactions can now be completed with one question.
Sure an LLM could have done this before, but not within seconds…….and it only costs $0.04 per million tokens which on average is about 80x less than an LLM.
Give it a go: https://Armchairexpertaiexplorer.com/
Subscribe