# I built an app where you ask about any moment in history and it turns it into a fully researched podcast you can interrupt — retained record

> Retained reddit record from reddit:ClaudeAI.

Canonical: https://highsignal.app/data/records/d5a522ed0e5c6a7c

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeAI)  published 2026-10-12 12:50 ingested 2026-10-13 02:37

[original source ↗](https://www.reddit.com/r/ClaudeAI/comments/1x40zxz/i_built_an_app_where_you_ask_about_any_moment_in/)

## retained text



```
I'm into history, and at work I do software and LLMs, so I spent a few months building the thing I wanted to exist: you type any topic, moment or person, and about a minute later two hosts are telling you the story, researched on the web with sources, with artwork that follows along. You can interrupt mid-episode with a question; they answer and pick the story back up. There's a quiz at the end if you want one.



How I built it

\- Claude Code did roughly 80% of the development. The front end came out of Claude Design.

\- I don't hand it a spec and walk away. I steer the idea and the design, it codes, I review, and we go back and forth, even on small features. Anything I'd regret shipping blind, I look at before it goes in.

\- I also used Sol/Astra to run Claude and GPT models against each other on the same problem. They have different strengths and often cover each other's blind spots, so letting them iterate on a design beat asking either one alone.

What's inside the app

It's not one model doing everything. I ran bake-offs across GPT, Claude and Gemini for each job: research and planning, script writing, artwork selection, voices, quiz. Some findings surprised me:

\- Opus 4.8 was the clear winner as the writer. Better dialogue, better pacing, and it handled "the sources disagree here" honestly instead of picking a side.

\- Opus 5.5, oddly, wrote more generic scripts in my tests. Newer wasn't better for this job.

\- Some models plan well but write flat; others are the reverse. Cost differences are large.

Learnings

\- Bake-offs beat intuition. My guesses about which model would win were wrong more than once.

\- Pin models per task, not per app.

\- Review is where the time goes, not typing.



I'd love feedback, both on the app and on how I'm working with Claude. You can try it at: [historai.ca](http://historai.ca)
```



document fetched 2026-10-13 00:23 · canonical https://reddit.com/r/ClaudeAI/comments/1x40zxz/i_built_an_app_where_you_ask_about_any_moment_in
