The Elon Podcast — AI-generated interviews with text-to-speech playback
How I built a podcast generator that uses chained AI models to craft interview scripts and produce lifelike audio — exploring the creative potential of multi-model orchestration while learning the front-end realities of generative AI.
The Elon Podcast — Podcast Generator with Text-to-Speech Audio Replay
Platform
Next.js · Vercel
Tech
Chutes AI · Qwen3 32B (LLM) · Kokoro (TTS)
My Role
Product Designer & Full-Stack AI Developer
Background
A creative experiment in chaining AI models to generate something people can actually listen to
Generative AI is often showcased through text and images — but audio is where it gets genuinely interesting. The Elon Podcast was built as a creative experiment to test what happens when you chain multiple AI models together into a single product experience: one model writes the script, another voices it.
The concept is simple but the execution is layered. The app frames Elon Musk as the interview subject, with the AI acting as an expert podcast host. Users type in any topic they'd like the interview to cover. The query is sent to Chutes AI, which uses Qwen3 32B to generate a natural-sounding podcast script, then passes that script to Kokoro to produce the text-to-speech audio.
This wasn't built for a specific audience or market need — it was a creative project designed to stress-test the Chutes platform's ability to chain multiple AI models into a cohesive product. The result was a working podcast generator that taught me as much about generative AI product design as it did about the technology itself.
What does the front-end experience of generative AI actually feel like for users — and what can a designer learn by building one end-to-end?
App Features
From a topic to a playable podcast in one flow
The Elon Podcast keeps the feature set tight — every element exists to move the user from a topic idea to a playable audio file with as little friction as possible.
Topic Input
Users describe any topic of conversation they want the podcast to explore. The input is open-ended — from technology and business to culture and philosophy. The AI adapts the interview to whatever subject is given.
AI Script Generation
Qwen3 32B crafts a full podcast script based on the user's topic, structured as a natural back-and-forth between the host and Elon. The script captures the tone and rhythm of a real interview.
Text-to-Speech Audio
Once the script is complete, Kokoro generates lifelike voice audio from the text — producing a playable podcast episode that sounds natural and conversational.
Chat Display & Audio Player
The output appears as a visual chat transcript between the Host and Elon, paired with an audio player for playback. Users can listen, pause, and download the generated episode.
UX Approach
How The Elon Podcast took shape
01
Simple Input, Rich Output
The interface uses a single text input — a pattern common across generative AI products like website builders and image generators. Users type a topic, press generate, and the AI handles the rest. The simplicity of the input belies the complexity of the multi-model pipeline behind it.
02
Building Full-Stack AI Intuition
This project was as much about learning as building. Working through the backend and frontend structure of a generative AI product gave me first-hand understanding of API latency, model chaining, and the UX implications of slow generation times.
03
Designing for Waiting
The biggest UX challenge was response time — AI generation takes longer than a typical API call. The current implementation uses progress indicators, but the experience highlighted the need for more engaging loading states that keep users informed and entertained during generation.
Product Design Patterns
Design patterns that guide users through a generative process
Generative AI products need to communicate clearly at every stage — especially when the process takes time. Three core patterns structure the experience.
Input
Topic text field
A clean, focused text input lets users describe their podcast topic. The single-field approach keeps the interface uncluttered and makes the first interaction obvious — type and generate.
Progress
Loading & progress states
The UI communicates each stage of the pipeline — from sending the query to Qwen for script generation, through to Kokoro for voice synthesis. Users always know where they are in the process.
States
Success & error handling
Successful generations load the chat transcript and audio player seamlessly. Server issues during any stage of the pipeline surface clear error messages rather than leaving users in ambiguity.
Playback
Audio player & download
Once generation is complete, a built-in audio player lets users listen immediately. A download button makes the podcast portable — users can save and share their generated episodes.
Tech Stack
A multi-model pipeline on a lean, modern stack
The Elon Podcast chains two AI models through a single API layer, keeping the architecture simple while the output stays rich and layered.
1
Next.js — Application Framework
Next.js provides the full-stack foundation — handling the front-end interface, API routes for the Chutes integration, and server-side rendering for the landing page.
2
Shadcn UI & Tailwind CSS — Design System
Shadcn components and Tailwind CSS provided a polished, consistent interface with minimal custom styling — letting the focus stay on the generative experience rather than visual overhead.
3
Chutes AI — Model Orchestration
Chutes AI serves as the API layer connecting both models. It handles the pipeline from receiving the user's topic to routing it through Qwen for script generation and Kokoro for voice synthesis.
4
Qwen3 32B & Kokoro — LLM & Text-to-Speech
Qwen3 32B is a 32.8 billion parameter language model that excels at reasoning, coding, math, multilingual tasks, and agent-based applications. It generates the interview script with natural conversational flow. Kokoro converts that script into lifelike speech audio — two models working in sequence to produce the final podcast.
5
GitHub & Vercel — Deployment
Version-controlled on GitHub and deployed to Vercel. The continuous deployment pipeline means every iteration is live within seconds of pushing to main.
User Journey
From topic idea to playable podcast in four steps
The user journey is intentionally linear — each step follows naturally from the last, with no branching or decision points to slow things down.
1
Enter a Topic
Users type the subject they want the podcast to cover into the text input. The prompt is open-ended — any topic works, from technology trends to hypothetical scenarios.
2
Generate the Podcast
Hitting the 'Generate Podcast' button sends the input to the AI pipeline. The UI reflects progress as the script is generated by Qwen and then passed to Kokoro for voice synthesis.
3
Listen to the Episode
The generated podcast appears as a chat-style transcript alongside an audio player. Users can read the script while listening — or just press play and sit back.
4
Download the Audio
A download button lets users save the generated podcast as an audio file. This makes it easy to share, archive, or listen offline.
Design Impact
What building a generative AI product taught me as a designer
The Elon Podcast was less about the final product and more about the process. Building a generative AI application end-to-end — from prompt design to audio output — gave me insights that directly improve how I design AI-powered products.
2
AI models chained in a single pipeline — script generation plus voice synthesis
E2E
Full end-to-end understanding of generative AI product UX — from input to audio output
Real
Hands-on experience with model switching, prompt tuning, and generation latency
The biggest takeaway was understanding the front-end experience of generative AI from the user's perspective. Knowing how long generation takes, what feedback users need during the wait, and how to handle errors gracefully — these are design insights you can only get by building the product yourself.
Takeaways
01
Expand beyond Elon Musk as a subject. The interview format works well — the next step is letting users choose from a range of public figures or create custom personas for the AI to interview.
02
Build a Character.ai-style personality interface. A gallery of different personalities — each with their own interview style and voice — would transform this from a single-subject experiment into a flexible creative tool.
03
Gather user feedback on the experience. The core generation loop works — now the priority is understanding what users enjoy most, what frustrates them, and where the product should evolve next.