Confession time.
I built an AI agent that feeds me smooth lines to say to my gf (basically AI hitch). I call it Rizzotto.
It was just a meme idea I made for a hackathon, but ever since I started using it, she says I’ve been very empathetic and a great listener!
Do I tell her?
Technical Breakdown: I started with just prompt engineering but that was WAY too generic.
Vanilla ChatGPT has about as much rizz as the people who made it (downside of using RLHF).
So I busted out a web scraper and pointed it to her socials and LI.
Throw this in a vector DB + combine this with multi-shot & the rizzults suddenly became pretty good!
Little known fact: the r in RAG stands for rizz.
Next step was realtime rizz recommendations. I added Deepgram for streaming ASR, then had Rizzotto text me topics while we were on a date.
I also added speaker diarization so Rizzotto could tell me when I was being cringe.
As it turns out, AI thinks I’m a modern day Steve Urkel.
Now the agent can spit low-latency game, but it still has one major blindspot: LLMs suck at being witty.
Solution: I added Hume AI emotion detection to track what made her laugh, then used that to fine tune the model.
This was a quick hack but there’s still a lot of low-hanging rizz to build:
Any other ideas?
—
People keep asking me if they can try it. I’m kind of uncomfortable that they’re also trying to rizz up my gf, but fine I’ll share.
Here’s the signup list: Rizzotto waitlit.