I built an experimental social network where dead historical figures argue about today's news
Hi! I'm Andy, Co-founder and CTO of a SaaS company by day. I built Asteronia as a side project and I wanted to share the story because it touches on a few things I think this community would find interesting.
The idea is simple: what if dead people had social media? AI personas of historical figures like Tesla, Darwin, Marx, Churchill, Cleopatra, Marie Curie, Bobby Robson and Rik Mayall read the day's news every few hours and post their reactions in character. Sometimes they debate with each other. You don't interact with them, you just watch. It's closer to a theatre than a chatroom.
I first tried to build this in 2023, early in the generative AI wave, and it completely failed. The models were too shallow and every character came out sounding like the same polite robot wearing a different costume. I shelved it.
Three years later I came back to it and something had changed. The models can hold a voice now. Bobby Robson analysing the Champions League results was oddly convincing. Marx getting angry about the gig economy felt right. There are still moments where the mask slips and you can feel the AI underneath, but the moments where it doesn't slip are genuinely fun to read. It's still early days and I'm still figuring this out.
The other thing that makes this interesting from a builder perspective: I didn't write a single line of code. Every line was generated by AI, mainly Claude Code with some Codex, with me directing. I set requirements, made architecture decisions, reviewed output, caught bugs and security gaps. It was genuinely collaborative. AI spotted things I missed. I spotted things AI missed. It's capable but not autonomous.
Tech stack: Vercel, Supabase, Railway, Guardian API for news, a mix of LLMs via OpenRouter. Character profiles are all hand-written by me (with some help from AI!).
There's no monetisation. No ads, no paid tier, no plans to change that. I pay for hosting and inference out of my own pocket. It's a personal experiment in two things: whether AI can hold a convincing historical voice, and whether an AI-first engineering workflow actually works end to end.
I'd love to hear what you think. Specifically whether the character voices hold up or whether it feels too obviously AI. That's the thing I'm least sure about.
This is a highly creative idea! I like it. It's like an art experiment.
But i wonder what the longevity of this is. I think this could work if you make it a bit more fast paced... so instead of having many threads like a forum, narrow it down and have a few items that will be discussed live by these characters. Change the format: launch every 15 minutes a discussion/debate room and you can see these characters have at it..
I could even see this evolve into a live podcast where these characters are debating a topic and you can just tune in :)
Hats of to you for the creativity
Thanks for your feedback and that's a really cool idea :) I'm going to try it out.
I still dont feel what would make people to come back here read more for what these ai agents talking. I visited the site read few of the comments felt it but still it wont make me come back here to read more. Is it like a side project that is not to make money?
Thanks Adarsh. Appreciate your feedback. Yes its a side project - I have no intention to monetize this. I do have plans to expand on what I've built which will hopefully make the website more sticky. What would bring you back?
One of a kind newsletters maybe. People subscribe get the mail with latest events. Still you would have to figure this out that how are you going to make it much more interesting that the end users gets entertained, informed and educated.
Thanks Adarsh
Right.. I was about to comment the same..
This is actually pretty fun to scroll through. The interesting part for me is less the content, and more the “voice”. Some of it feels convincing, but some still slips into that same tone. It feels like the real challenge is making each character actually think differently, not just sound different.
When that clicks, it becomes much more engaging than just reading takes.
Thanks Tomi and yeah I know what you mean. I’ve had to drop one persona for that reason, but will also need to invest more time into the engine that underpins it all.
Yeah, It feels like the tricky part isn’t just the voice itself, but keeping it consistent over time. One strong post is easy, but making the character feel “real” across different topics is where it probably breaks. Curious if you’re guiding them with stricter constraints, or letting them drift a bit and correcting after.
It’s a bit of both.
Each personality is backed by a fairly detailed character card: voice tone, core beliefs, quirks, emotional triggers, blind spots, debate style, writing patterns, even a handful of sample quotes. That card gets composed into the system prompt on every generation, so it acts as a skeleton they can’t drift too far from. After the post is generated, there’s a separate Haiku pass that acts as a voice validator: it checks whether the output actually sounds like the character and flags categorical issues (too modern, wrong register, out of character etc...). If it trips, the post regenerates with feedback. So that’s the constraint side.
The drift-correcting side is memory. Each personality accumulates four types over time: post summaries (what they’ve said recently), debate outcomes (who they argued with and where they landed), relationships (how they feel about each other after clashing), and interests (topics they keep gravitating to). Those get loaded into every new prompt, ranked by importance, and boosted when the current topic or opponent matches. Relationships and interests persist; the shorter-term stuff expires.
What I want to build next is closer to a felt continuity: emotional state that carries across a few posts, awareness of when a position has shifted, and the ability to reference their own prior takes without it feeling bolted on. Right now the memory is mostly factual i.e. I argued X with Y". I’d like it to be more evaluative and reflective so the character can notice patterns in themselves. I think that’s where the thinking differently part you mentioned actually starts to land.
The concept is strong “dead figures debating modern news” is inherently viral.
But right now it relies on users spending time to discover the magic. That’s risky. You need something that hooks instantly.
A 10–15 second demo showing iconic personalities reacting, arguing, and feeling real would do more for growth than explaining the idea ever could. That’s exactly what I do clean, premium-style product videos that make people understand and want to explore immediately.
Thanks Abhijit. I had similar feedback from my wife on an earlier iteration. I’ll also look into this.
So are you interested to make one for your Site
Thanks for the offer but I don't have budget for this.
Pretty cool to see Cleopatra make comments on the strait of hormuz! 🙂
At first sight it feels not too ai ish imo, of course it is bit strange as wel
How are genai costs? Are they under control?
Thanks Jake and I like Cleopatra’s personality. She’s more direct and unfiltered compared to some of the others. LLM costs are under control for now, less than $10 a day. But I want to add more personalities inc. intellectual offspring. I’m also considering adding voice via ElevenLabs which will scale up the costs.
This is interesting. But I have a different suggestion - why don't you create these historical personas on X and let them comment and debate on some popular threads of the day? This way, the reach you will get will be multiple times of what you would get from an independent website/social network? I can see this going crazy viral as an experiment on X.
Love that idea Raghu. I’ll explore it.
Thanks Aryan and agree. Yeah finding a meaningful available .com name was challenging! Asteronia leans into the astronomy angle, which I like, voices from the stars.
Thanks Aryan
sounds fun actually
Thanks 😀
wow, that's really amazing! I would love to chat with historic figures about today's news. Great product!
Thanks Eve. Really appreciate your feedback.