I'm excited to show off this really fun project I put together. I originally built this project 2-3 years ago, AI was already booming at the time, however voice AI agents were still very early. I loved my proof of concept at the time, but wasn't quite happy with it.
I recently had the desire to check out the tech again, and know many of you will be interested.
Interviews are speech to speech with OpenAI's gpt-realtime-2.1 over WebRTC. This model is... expensive, and because of that, I have to add some amount of restrictions, conversations are tied to a authenticated Clerk user id. I have also added a 30 minute timer because well, I really don't want to go broke while I sleep tonight.
Each suspect has a tool they call when you make a direct accusation. It captures who you accused and a faithful list of the evidence you actually stated.
A separate gpt-5-mini judge then decides which of the case's required evidence facts you genuinely presented. Paraphrasing counts, vague suspicion and fishing don't.
The rest is Next.js, MongoDB, and Clerk.
Let me know whether the suspects hold up under a real interrogation.
I've always thought that a game like Civilization should allow the user to interact with different opposing world leaders, but in a non-deterministic way as opposed to a subset of fixed interactions. Given what you know about this leader's temperament, goals, and so on, you can conduct diplomacy in a way that can be way more immersive and crafty than what the game currently gives you, which hasn't seem to evolve much in decades.
I've been working on a similar concept for 6 months now (https://game.selna.ai, portuguese only for now, sorry!), and it's been a journey. I've been working with VR professionally for some 8 years now and creating what I call a "social virtual reality" seemed simple enough, but has turned out to be much more complex than I initially imagined.
In the same way that in normal VR a single badly scaled object can break immersion, in "social VR" a single contradiction kills the experience. But when stars align and eveything works, it's magical.
I started with whodunnits but have expanded it into linear and non-linear storytelling. At this point the generation pipeline has some 40+ separate steps, the output of ones feeding others, until it all comes down to single prompt the NPC uses to talk using the realtime API. The platform can create a complete narrative (including cast personalities and images) in some 20 minutes, and in the process generates the equivalent to a big book of text.
I've been working on this just because I can and I am a bit obsessed with getting it working perfectly, and I don't really know what I'll do with it when it does.
I think this would work very very well in combination with human writers.
The developer's homepage gives the impression that this is more for the tech and fun aspect, rather than collecting speech samples.
what a strange comment
Although most time I cave in, I really understand the sentiment of the parent.
Did the server go down or did I hit one of your limits?
As a second point - this is very interesting using voice control as game input are there viable alternatives for realtime voice uasing your own GPU?
I launched this at midnight prior to logging off and going to bed, funnily enough a massive storm hit us around 3am and I woke up and noticed my OpenAI account failed to fund lol. So we ran out of funds and it caused things to break for some of you.
Just waking up here on the East coast, once my daughter is occupied I will try to answer more of your comments!