1:1
One match at a time.
An AI matchmaker rather than a marketplace: it learns who you are in open conversation, then introduces you to exactly one person — three photos, a first name, and a written case for why her. Live face verification makes one person, one profile a constraint, not a rule.
// the_objective
Build a dating product that spends attention rather than harvesting it. No swiping, no browsing, no feed — an AI matchmaker that learns a person in open conversation, keeps only what they agree to keep, and introduces them to exactly one other person at a time, with a written case for why.
// our_approach
Our approach
Conversation instead of a form
There is no questionnaire. Two matchmaker personas — Sam, who listens, and Sky, who will roast you — learn a member the way a friend does, and a five-dimension personality portrait is inferred from how someone talks rather than asked for. Switch persona whenever you like; both share the same memory of you.
Consent-gated memory
Every extraction surfaces in the thread as a quiet "just learned" line beside the member's own words, which they can leave shared, mark private, or delete outright. Memory is an interface element rather than a hidden profile — nothing sticks without an explicit decision.
Identity as physics, not policy
Onboarding requires a live selfie: a liveness check that it is a real capture and not a photo of a photo, a match against the member's own pictures, and a search against every verified account. The resulting face vector lives in a private index for the life of the account, so one face can only ever hold one profile.
One introduction, fully argued
When the matchmaker is sure, the app changes shape: three photos, a first name, and a brief in full sentences under three headings — what you share, where you balance, and be careful. Introductions run in scheduled rounds, and nobody is queued behind the person you meet.
A stand-in that never pretends
When one side steps away, their matchmaker holds the thread under an explicit "via Sam" label, sharing only what its person agreed to share and deferring anything only she should answer. Voice notes carry tone with a word-for-word transcript on tap, and translation runs both ways so each side reads and writes in their own language.
// the_product
The product
// key_highlights
Key highlights
Nobody can browse you
No feed to appear in, no search to be found by, no grid of faces to be judged from. The only person who ever sees a profile is the one chosen for it — which makes a whole class of unwanted contact structurally impossible rather than merely against the rules.
Routed models, no training on members
Reasoning is routed across several providers through a single gateway, configured so member conversations are not used to train them — while speech-to-text and face verification run as separate, single-purpose processors.
Deletion that actually deletes
One path inside the app removes the profile, photos, verification selfie, biometric face vector, every matchmaker and match conversation including voice notes, the personality profile and the extracted learnings — completing within 30 days.
Built to the store bar before the store
Eighteen-plus with no exceptions, human-reviewed reports with the most serious surfaced first, instant and permanent blocking, and published community and child-safety standards — all in place ahead of the iOS and Android launch.
// the_stack
Tech stack
Convex
Application backend, data and realtime sync
OpenRouter
Model routing, configured not to train on member data
AWS Rekognition
Liveness, photo match and duplicate-face search
Deepgram
Speech-to-text for voice notes and voice input
Authkey
WhatsApp one-time codes — sign-in without SMS
Cloudflare
Media storage, served through short-lived signed links
Astro
The pre-launch site: two JS bundles, no trackers
PostHog + Sentry + FCM
Product analytics, crash reporting and push
// faq
Frequently asked questions
- Does Foways build its own products, or only client software?
- Both, and products come first. Foways is a product studio: we design, engineer and run our own software end to end, and we take a small number of custom briefs each year alongside it. Every line is written by our own senior team, never subcontracted.
- Can a product be deployed for a single department or organisation?
- Yes. Setu and Foways Office are both delivered as single-tenant deployments — one instance per customer, self-hosted on infrastructure you choose, updated over the air by us. Nothing is shared with another tenant.
- Can Foways adapt a product to our own requirements?
- Yes. Both government products are built to be extended rather than configured around: capability is added in TypeScript through a stable module interface, so a department gets what it actually needs without forking the product or losing its update path.
// next_step
Talk to us about 1:1.
Tell us what you're trying to run. We'll show you the product working, scope the deployment, and give you a straight answer on timeline.