Building @heywavelength | Member @MensaInternatl (high iq society) l
india
Joined April 2024
- Tweets1.2K
- Following1.3K
- Followers791
- Likes4.5K
A song can't reject you back. That line is the whole problem.
Three things that make matching people harder than matching music:
1) The signal is voice. Someone talks to our AI for 20 minutes. From that we extract :
- who they are
- what they actually want
- and how sure we are about each
Confidence is a first class number in the scoring, not an afterthought.
2) Both sides have to say yes. Every score is two scores. A match only exists if both directions clear.
3) Taste drifts. Every call, message, reveal and unmatch is new evidence. The model of a person has to keep updating without overreacting to one bad date. It's a full feedback loop, not a one time profile.
Spotify solved taste in music. TikTok solved taste in content. Nobody has solved taste in people. Not Tinder, not Hinge, not anyone.
If you think you're smart enough, DM me or fill the form in the comments
We're building the world's most advanced human compatibility graph.
Tiktok cracked taste in content, Spotify in music. Nobody has cracked taste in people.
It's a harder problem. A song can't reject you back. A match only works if both people say yes. Listed a few problems below:
1. Using voice to understand someone's true intent and personality.
2. Bi directional matching.
3. How do we keep updating our understanding of someone's taste as context grows.
4. Memory, Evals, Cost, etc.
Hiring primarily for ML + backend.
In office HSR, Bangalore.
Fill the form below or best if someone can refer you. Or DM Jayanth.
binary.so/V24XmYl
Link to the form : binary.so/V24XmYl
1
4
122
hiring for a cracked video editor at @heywavelength
how does one meet cracked video editors who have a great music taste and are ultra passionate about creating content.
over the next few months we're gonna attempt to completely reimagine how content marketing is done in india.
we're building team of passionate content folks - if you'd want to know more, pls ask someone we know to refer you.
or fill the form: forms.gle/jeBxFbBM77qUVmcD6
mascot FTW. This can be a crazy customer delight
The first version of our mascot looked like an infected pimple.
Not a joke for engagement. I have the file.
Some context: I'm the founding designer at Wavelength, a dating app built around Wave, a companion you actually talk to. It listens to your most unfiltered thoughts and finds you people. Something like that can't be a chat bubble. It needed a body.
Here's every way I tried to make one before it worked.
Attempt 1: vibe coding. Me and Claude in VS Code, @DramaticKing__ patiently explaining push and pull requests to me like I was a small confused child. We asked for a soft, glowing, living blob. We got an infected pimple.
Attempt 2: Unicorn Studio. Genuinely beautiful. Also ate the phone's GPU alive the moment it hit code. A companion that drains your battery isn't a companion, it's a parasite.
Attempts 3 and 4: Rive's agent, then Rive's MCP through Cursor. Surely the AI tooling could just make the thing. It could not.
The first honest lesson of the project: you can automate a shape, you cannot automate a soul. Every generated version was technically a blob and emotionally furniture. So I opened the Rive editor and built it by hand.
About a year ago, Microsoft's CPO Aparna Chennapragada called NLX, natural language experience, the new UX. Conversation becoming the interface itself. Wavelength lives that idea completely, no swipe grid, no filters, the whole app is you talking to Wave. But building one of these teaches you something fast: a conversation-shaped interface still needs things you can see. Nothing says "I'm listening." Nothing says "got it." Most apps hand those jobs to a spinner. We handed them to a body.
So the brief was never "make a cute mascot." It was: give the conversation a face.
Wave's body is a few soft layered circles and two eyes. The restraint is deliberate, nobody wants a cartoon doing backflips while they admit they got ghosted.
The breakthrough came mid-build, on a nod animation of all things. I was about to move the whole orb like a rigid puppet head when it hit me: the centre circle isn't a layer. It's the soul. The soul moves first, the body follows. That rule now runs through everything Wave does. It breathes when it's present, takes a slow intake when it's listening, dips into a small nod when it's heard you. And yes, it side-eyes. Tell it you're texting your ex and watch.
Not everything landed. I knew Rive, but data binding, the part that lets Wave react to you live, was new territory. Chetan and I figured it out together. And in the first build, the blink didn't come through. Wave stared, unblinking, into the void.
Wave isn't fully out in the wild yet. Something fun is coming.
And there's a part 2, because your Wave and my Wave won't even be the same colour.
(Also no, we're not changing the logo. The team did ask.)
4
1,025
I've been building Wave for some time now and have had countless conversations with her.
Three weeks ago I mentioned a restaurant I wanted to try in passing, and it remembered
It took me five minutes to even remember saying it. It made me feel a warm, almost humane connection just because it remembered a small thing.
While most AI agents suffer from digital amnesia, where every conversation starts from zero, what if Wave can remember things about you just like a good friend would?
Because we know that trust isn't built through grand gestures. It's built through a hundred small moments where someone proves that they were listening.
That's what we're trying to build with Wavelength. Full piece in comments.
see you at inception soon folks
super excited to have @ejayO9, CTO and co-founder of @heywavelength as a judge at Inception.
india's first world model hackathon.
june 20, bangalore.
Claude Fable is crazy....
I asked it if it thinks i am successful. It then started looking at my product's posthog data to see if i am successful .
Nobody told me building an AI dating app meant becoming a salesperson.
Matchmaking is a two-part problem . Part 1 is the technical part. Our AI conducts hours of deep context interviews, analyzes data, and honestly knows these people better than their therapists and moms combined. We have substantial proof that User A and User B should be arguing about what pizza to order because they both like different pizzas (told you we did our research)
Then comes Part Two: The Pitch.
Turns out, humans are not rational animals. You don’t look at a perfectly curated, data-validated match and go:
"Ah yes, the compatibility is 94% . I shall now ask her for a date."
No. Your brain sees a photo, notices they are wearing a specific shade of yellow that reminds you of the teacher that scolded you 15 years ago, and goes: "Idk man, vibes are off."
So we have to sell you the person.
And here's the fun part (nightmare for us to code): everyone needs to be convinced differently. The pitch that works on you will completely fail on the next person. Not because people are complicated in some vague, poetic way, but because your preferences have origins you're not fully aware of.
You want someone calm. Fine. But is that a genuine preference or is it what you want after years of loving someone who wasn't? Those are not the same. One is a type. The other is a pattern you're still carrying. We need to know which one we're working with before we find someone for you. So we don't just send a profile. We send a pitch for the version of you that actually knows why you want someone calm, not the version that just filled out a form.
We don't say, "Here is Sarah." We say, "Hey, look. You told us you hate loud chewers and need emotional stability. Here is 4K footage of Sarah eating cereal in a crisis. Trust us."
Last match we ran, beta users loved it. The WhatsApp feedback group didn't immediately roast us, which in startup terms is a massive victory.
A matchmaker's job isn't just finding the right person. It's being the friend who pulls you aside and instead of saying “Trust me bro” says “ I’ve been paying attention when you weren’t”
That’s the job.
@heywavelength
we are hiring for founding engineer roles folks. do apply if you think you have crazy problem solving skills
Wavelength is hiring a Founding Engineer via Backdoor.
If you're an engineer who loves building, shipping, and taking ownership from day one, this is for you.
Text Backdoor to apply, link in next tweet
@heywavelength @vinitsarode_ @ejayO9
at @heywavelength you can build your partner
1
6
530
turns out finding good men in bangalore is exactly like pokémon. rare spawns
you have to know where to look
we looked
the pokédex is live on @heywavelength . go catch yours.
4
312
@heywavelength finds you a date. @justswishin makes sure you don't show up hangry.
We do the matching, they do the cooking. Priorities sorted. 🤝
7
213