Discovery is abundant.
Belonging is scarce.
Circles is positioned as a living map of people, communities and plans already overlapping around you. The goal is to turn one-off interactions into familiar faces and repeat connection.
“Who do I go with?”
Even in a city full of activities, the effort of finding company, coordinating and showing up can stop a plan before it starts.
Plans as a social bridge
Shared context and concrete activities may make introductions feel warmer and lower the barrier to meeting offline.
Repeat connection
Measure whether people reconnect, not just whether they browse, match, or attend once. Proposed metric; not a reported result.
Positioning adapted from the public Circles homepage.
Four acts, one relationship loop.
These four steps are described on the public Circles website. The design rationale below is my product interpretation; the public page does not establish funnel performance.
Express your world
Neighborhood, interests and communities give people context beyond a static bio.
Find warm signals
Shared interests, friends, communities and plans create a more natural starting point.
Make it concrete
Curated nearby plans move users from abstract intent to something they can attend.
See each other again
The goal is repeated familiarity, not an endless supply of new introductions.
Designed around belonging, not endless swiping.What we optimize for
Relevance over volume. A small set of credible plans instead of overwhelming choice.
Context over cold starts. Social overlap gives people a reason to connect.
Offline outcomes over screen time. The desired behavior happens outside the app.
These are positioning and design principles, not claims of measured lift.
A little more familiar.
People you
overlap with
Communities for you
The slow Sunday crowd
Matcha · making things
WHAT CONNECTS YOU
You & Maya
MayaSame chapter. A few things already in common.
FRIDAY · 7 PMEVENTS FOR YOU
Sunset picnic
at Pier 45
Hudson River Park
What connects you · a low-key evening
Wait, hi again!!
Love that we’re reunited ♡
Make the reason visible.
A stranger’s profile tells you about them—not what connects you.
The choicePut shared interests inside the intersection. Show concrete common ground, not a compatibility percentage.
A simpler explanation leaves out nuance, but gives people something real to talk about.
What I would measureProfile viewed → shared item noticed → interest in meeting.
Matching is a product decision, not a magic score.
Circles' founder reports using structured matching signals and replacing unreliable LLM-written match explanations with deterministic templates. This portfolio intentionally omits the production weights, scoring function, thresholds and group-formation algorithm.
Illustrative logical stages only. No claim is made that these are the exact live services, sequence, or safeguards.
Explore the matching tradeoffs
Move the sliders to see how different signals can affect a simple public-safe suitability illustration. Equal weighting is used solely for this demo.
Explore group fit
Not a production score or prediction. Real matching requires consent, safety, context, group composition and evaluation beyond this toy example.
Why we reduced LLM-generated explanations
Observed problem (founder-reported): Generated matching copy sometimes asserted shared traits unsupported by source data.
Decision: Use deterministic, data-grounded explanation templates rather than free-form generated claims.
Tradeoff: Less expressive language, but stronger factual control, inspectability and user trust.
How I'd evaluate it: Unsupported-claim rate, user comprehension, complaint rate, match acceptance and downstream attendance.
AI judgment means knowing when not to use generative AI.
From product idea to working application
I built Circles using Lovable, applying AI-assisted development to iterate on product experiences and workflows. This demonstrates hands-on prototyping and an ability to connect UX requirements with implementation. The public website alone does not verify the full production stack, security controls, integrations or app-level adoption; those details are deliberately not claimed here.
Strong foundations. Deliberate boundaries.
I used AI-assisted development to build the product. The important work was deciding what the system should know, guarantee, and never assume.
Experience
React · TypeScriptPeople, overlap, circles & plans
Product logic
TanStack server functionsMatching, member actions & verified webhooks
Data & access
Postgres · access policiesProfiles, memberships, messages & payments
Make overlap a data relationship, not an AI verdict.
Interests, intent and life context are explicit signals. Shared-item explanations come from stored answers, so a person can see the actual starting point.
Structured choices miss some nuance. A ranking is a useful introduction—not a prediction of friendship.
- Implemented evidence
- Profile taxonomy → overlap services → connection cards
- Acceptance check / next test
- Can a guest explain what connects them to someone without seeing a score?
My AI thesisAutomate the work around connection—not the human judgment of who deserves one.
Product principle, not a research findingTools accelerate execution. They don’t own the decisions.
Lovable supported the build. Managed infrastructure provides auth and storage; Stripe, Razorpay and messaging integrations appear in the codebase. Their presence is implementation evidence, not a claim that every provider is live or that the entire system has passed a security audit.
A relationship product with an offline proof point.
Early paid-event validation supports a narrow conclusion: people were willing to pay for a curated social experience. It does not establish app retention, repeat engagement or scalable unit economics.
Positioning against alternatives
| Approach | Primary job to be done | Circles' intended differentiation |
|---|---|---|
| Event discovery platforms | Find something to attend | Surface the people and communities around the plan |
| Friend discovery apps | Meet potential friends | Anchor introductions in shared real-world activity |
| Curated social experiences | Attend a facilitated gathering | Build continuity beyond a single experience |
Qualitative positioning framework, not an independently validated competitor feature audit.
Start with high-intent experiences
Curated gatherings can test willingness to pay and the operational requirements for quality. Possible longer-term models include ticketing or venue partnerships, but these are not presented as proven recurring revenue streams.
Density before breadth
A neighborhood-first launch can help concentrate plans and community overlap. The tradeoff is slower geographic expansion in exchange for a stronger chance of repeat interaction.
What I'd test next.
These are proposed experiments, not shipped features or achieved outcomes.
Measure the second encounter
Instrument opt-in post-event feedback and repeat co-attendance. Test whether continuity prompts improve meaningful reconnection without feeling intrusive.
Primary metric: repeat meaningful connection within 30 days.
Improve first-plan activation
Compare a focused set of context-rich plans with a broader browse experience, while holding neighborhood supply constant.
Primary metric: qualified plan attendance per activated user.
Earn trust in matching
Test whether clear, grounded explanations improve users' confidence and attendance relative to opaque recommendations.
Guardrail: misleading-explanation reports and opt-outs.
Product philosophy
Start with a real behavior, not a fashionable technology. Make the tradeoffs explicit. Ship the smallest credible experience. Measure what happens in the world.