01What Jev AI Is and How It Works
Jev AI is a decision-making model that picks an answer from a list in about half a second. Unlike most AI models that write full sentences, Jev never writes at all. You give it three things: a situation, a question, and a list of allowed answers.
It picks one answer and tells you how sure it is with a percentage, like 81 percent for a red blazer.
Jev is built for snap decisions, like stopping at a red light. It costs 95 to 99.5 percent less than the models it was tested against. Because it’s so fast and cheap, you can ask hundreds of small questions without noticing the cost or delay.
02Why Jev AI Makes Real-Time Apps Possible
Three features make Jev perfect for a live app like the outfit changer described in this episode. Speed is the first reason: a mirror that takes five seconds to react feels broken, but Jev answers faster than you notice you asked. A traditional writing model took 4.6 seconds for the same decision.
The second feature is constrained answers. A writing model might suggest a jacket you don’t own, breaking the app. Jev can only choose from the closet list you give it, so every answer is something the app can actually use.
The third feature is batch questions: one sentence can be split into seven questions at once, like what jacket, shirt, trousers, shoes, extras, and room to use. All answers come back in under a second.
03How to Build a Real-Time App with Jev AI
- Set up the ear: Use Chrome’s built-in speech-to-text to turn your voice into text. You can also type instead. No AI is needed for this part.
- Create the brain: Run a small program on your Mac that sends Jeff your sentence when text arrives. Include the current wearing, seven questions with allowed answers, and a closet list. Each item has a name and a one-line description of when to wear it, like “red blazer: Julian’s signature on stage look for keynotes and conference talks.”
- Add a quality check: After Jev picks, ask it a second question: how well the outfit fits the request on a scale of 0 to 3. If the score is under 2, tell Jev the look clashes and let it pick again.
- Build the tailor: Use two modes. Photo mode takes one photo of you and repaints only the clothes and room. The first look takes about 10 seconds, then saved looks appear instantly. Live camera mode uses your webcam and a real-time video API called Lucy. Each time Jev decides, the app sends Lucy a sentence like “change alpha to red blazer, white t-shirt, navy trousers” and Lucy changes the moving picture in 1 to 2 seconds.
- Glue everything together: Put one web page and one small server between them. Keep your API keys on the server so nothing secret reaches the browser. All intelligence is the list of questions and the one-line description per item. Everything else is plumbing.
04Using Jev AI: Walking Through the Outfit App Example
In this episode, Julian Goldie demonstrates a real-time outfit changer app built with Jev AI. He starts by wearing a t-shirt and shorts and speaks into the microphone: “I’m giving a keynote tomorrow, which jacket should I wear?” The app listens, sends the sentence to Jev, and Jev picks from a closet list that includes “red blazer,” “leather jacket,” “overcoat,” and “keep what I have.”
Jev returns “red blazer” with 81 percent confidence. The quality check scores the outfit at 2.5, so the app sends a command to Lucy: “change alpha to red blazer, white t-shirt, navy trousers.” In about one second, the live video shows Julian wearing the new outfit. He can also change the room background by saying “put me on a conference stage.” The entire loop—speak, decide, change—happens in under two seconds.
05Cost and Speed Test Results with Jev AI
Julian compared Jev AI against Haiku and Sonnet 5 models. Jev was 95 to 99.5 percent cheaper for the same decisions. The cost display in the app showed very low numbers per query. Speed was also dramatically faster: a traditional writing model took 4.6 seconds to answer one question, while Jev answered in under half a second.
That speed is what makes the live video feel instant instead of laggy.
Julian also tested building the app using Claude (Fable 5.1). The AI coding agent created an image version of the app that ran on localhost, but it was not as polished. The live camera version required a combination of Fable and Jev to work smoothly.
The heavy coding was done entirely by the AI agent, showing that non-coders can create similar apps.
06Tips and Common Mistakes to Avoid with Jev AI
Write clear decision lines: If your app does something silly, the fix is almost always a clearer description of when each answer fits, not a code change. Always include a “do nothing” or “keep it” answer as an option. Use the confidence score: When Jev’s confidence is high, let the app act. When it is low, have the app ask you for clarification or do nothing.
This prevents the app from making bad choices and breaking.
Pick a fast output API: For this app, Julian used Lucy because it changes video in 1 to 2 seconds. Testing with real sentences during development is crucial. You can watch the cost per query in real time, which helps you tune your app to stay cheap.
Also, because Jev never writes paragraphs, it cannot explain itself—so you must design your questions and answers carefully.
07Who Should Use Jev AI and Why It Matters
Jev AI is for anyone who wants to build real-time interactive apps that react instantly. You do not need to code; you can tell a coding agent like Claude what you want in one sentence, and it builds the app for you. The real skill is writing the decisions: the questions and allowed answers in plain English with one line of when each fits.
This approach matters because it makes AI apps practical for everyday use. A virtual try-on, a voice-controlled slideshow, an interactive shop window, or a game that generates itself step by step—all become possible with very low cost and near-zero delay. Julian mentions that the same system can power any app where a fast, constrained decision is needed, opening up new possibilities for creators and businesses.
