Hey HN!
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.
Good way to get more people's voices into training dataset
Neat! I've been wanting to make something like this for a long time, well before LLMs came around. It was one of the first things I tried out with an LLM in the early days of ChatGPT. My son and I made little mysteries for each other and just pasted them in a chatgpt conversation with instructions for it to pretend to be the other characters and whatnot, then handed the keyboard to the player. Mine didn't work that well, I don't think he ever solved it; partially because chatgpt kinda led him astray. When I started his, I just point blank asked the mystery and it told me the solution. I felt bad about that, he had put a good amount of effort into it, but I had reminded him to tell it not to reveal the answer and he didn't. Anyway, I figured someone would make this eventually! Excited to give it a try!
sounds good, but it won't get past Dispatch confirmation. You should specify what networking permissions or whatever is needed. My It goes red and the text to read disappears.
Same.
Vibecoding at its finest
This is a really fun take on a detective game. One small thing I noticed: during the initial detective briefing, the time always seems to show 0:00. Not sure if that’s intentional or a bug.
Maybe it only works for hn accounts created yesterday
Even the tag-line, "one poisoned patriarch, one locked-in truth" sounds very AI, and makes me think that I'd struggle to enjoy this. But I'll give it a go, because the idea is excellent.
I think this would work very very well in combination with human writers.
even without trying it, its so cool this exists. I think this kind of voice interaction has been a dream of adventure game creators for so long, and now its possible. there was a game many years ago (in the days of messengers like trillium being popular), I forget the name, where one of the characters would make a landline phone call to you. The voice acting was great and it felt novel and exciting to see it work.
It was working great... But now it says it can't connect.
Did the server go down or did I hit one of your limits?
sounds neat, would love to try it but I don't make accounts for websites anymore.
so if you got locked out of hn, you would just never comment again?
what a strange comment
I would assume it implies “websites that have no need to actually have an account”.
Although most time I cave in, I really understand the sentiment of the parent.
It makes perfect sense. You are being intentionally obtuse.
[flagged]
"voice connection failed"
same here. js error though is 0oxvw2ytjke1_.js:1 POST https://api.openai.com/v1/realtime/calls 429 (Too Many Requests). and auth bearer token in clear js request here towards openai. someone could have just used the token for their own good. cant go directly to openai's apis from the website. huge security risk.
True, I have a dungeons and dragons combat tracker (https://topoftheround.com) where I let the user generate loot items with different settings for them like rarity or magical/non-magical etc. and I am using a firebase cloud function that talks to the actual AI and returns the result. I also have set a limit to how much that token is allowed to spent per month, hopefully OP has thought of that as well. Interpreting and generating voice probably uses a lot of tokens.
[flagged]
[flagged]
I have no problem with AI generated text and story but I just can't stand AI voices. Even a hint it is AI narrated and I check out. I would encourage you to consider getting it voice narrated by real people. I'd rather listen to bad/amateur human narration than AI voices.
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?
The entire game hinges on the premise of being able to narrate and understand dynamic text on demand. Pre-recorded messages don’t work for this.