I built Zygnal because I kept seeing the same thing happen to people (and honestly, sometimes to myself too):

You open a dating app, upload some photos, write a bio, hit publish… and then you’re basically doing blind experimentation in public.

If it works: great, you assume you’re “good at dating apps.”
If it doesn’t: you start spiraling.

You swap photo #1. Nothing changes.
You rewrite the bio. Still nothing.
You ask friends and they’re either too nice or wildly unrepresentative.
You try online advice and it’s either generic (“smile more”) or toxic (“neg women”).
After a while, the most common outcome isn’t “better profile” — it’s less confidence and more burnout.

That felt unnecessary. Like… a lot of suffering caused by a system that gives you no feedback loop.

So Zygnal is our attempt to add the missing piece: measurement + guidance without dehumanizing people.

What we believe (core values, in plain English)

People aren’t products. We don’t want “hot-or-not scoring.”
Truth over tactics. We’re not building pickup tricks. We’re building clarity.
Compatibility > mass appeal. The goal isn’t to be liked by everyone, it’s to be legible to the right people.
Evidence before narrative. AI should summarize reality, not invent it.
Less suffering. If we do this right, people spend less time guessing and more time actually connecting.

What Zygnal actually does (the engineering thesis)

A dating profile is a tiny compressed signal: 6 photos + a few lines.
Other humans do fast inference from that signal. If the signal is noisy, they don’t “reject you” — they just… hesitate. And hesitation kills you in swipe feeds.

So we model profiles as latent variable systems: there’s an underlying “vibe / compatibility potential,” and your photos + bio are observations. The market’s reactions are noisy measurements.

Instead of a binary “good/bad,” we collect ordinal crowd feedback (because human judgment isn’t binary), then infer a posterior using an ordinal Bayesian model we call VCI (Vibe Compatibility Index) — explicitly designed to avoid objectification and focus on audience resonance.

We also correct for demographic skew using MRP-style adjustments, because “internet average” is often a lie. What lands with one segment can fail in another. Zygnal can show segment truth: who you resonate with, where uncertainty is high, where you’re misread.

Then we split the pipeline into two parts:

Deterministic diagnostics (evidence):
- photo attribution signals: clarity, trust, authenticity cues, redundancy, context quality
- portfolio coherence: do your photos tell one readable story or conflicting stories
- bio flags: negativity/self-sabotage, vagueness, mismatch between stated intent and vibe
LLM coaching (but grounded):
We use targeted prompts to turn the evidence into a plan you can execute — but we constrain the LLM to the measured findings so it doesn’t hallucinate “dating guru” nonsense.

So the output isn’t “be hotter.” It’s stuff like:

“photo #1 creates uncertainty (eyes/lighting) → swap to identity-lock shot”
“you’re over-curated → add one candid moment to reduce trust debt”
“you attract volume but not the audience you want → adjust intent signals”
“this segment reacts strongly; this segment is indifferent”
plus concrete, actionable photo/bio changes

Where we’re going (ambitious, but honest)

Today we’re focused on profiles — the first bottleneck.
Long-term, the real world is dynamic: people aren’t static profiles. So we’re exploring (carefully, responsibly) how to model state and interaction patterns — not as “love prediction,” but as a way to understand compatibility trajectories instead of one-time impressions.

Stack (for the builders here)

Built with @Expo, backend on @Google Cloud, and we use Gemini for the narrative layer (with guardrails + evidence grounding).
#expo #reactnative #googlecloud #gemini #ai #ml #bayesian #statistics

If you’ve ever felt that “I’m doing everything and still getting nothing” frustration on dating apps — that’s the exact pain we’re trying to reduce. And if you’re a builder who cares about measurement, feedback loops, and human-centered AI, I’d genuinely love your feedback.

Zygnal is built on these non‑negotiables:

Dating apps are markets.
- People decide quickly using thin‑slice heuristics.
- The profile is a compressed signal packet.
Noise is guaranteed.
- Raters differ in taste and strictness.
- Markets differ by region/age/platform.
- Preferences drift over time.
Truth requires inference, not opinions.
- We can’t read “true attractiveness” directly; we estimate it from noisy observations.
Optimization must be segmented.
- The goal isn’t to appeal to everyone; it’s to appeal strongly to the right people.
Action must be measurable.
- Zygnal must convert diagnosis into an intervention and prove improvement.

What Zygnal is (and isn’t)

4.1 Zygnal is

A measurement system for attraction signals under uncertainty.
A marketplace that pays for attention with attention.
A decision assistant that outputs prioritized improvements.
A retest loop that validates uplift.

4.2 Zygnal is not

A magic “hotness score.”
A clinical psychological diagnostic tool.
A guarantee of dates or relationships.
A replacement for personal values or consent.

4.3 Product promise (safe + defensible)

Zygnal helps users reduce uncertainty about how their profile signals are received in their target market, and guides improvements that measurably increase profile performance.