From a no-code Softr stack to a custom matchmaking platform with AI support
Matched By Us is a custom platform I built for a premium matchmaking agency in Flanders. It replaces a no-code stack with a SvelteKit application on Cloudflare, including human-supervised AI matching, customer portals, contracts, planning, payments and communication. More than 157,000 lines of application code, 39 database tables and 4,668 automated tests were delivered in approximately six months.
The challenge.
Matched By Us is a premium matchmaking agency with eight matchmakers. Its operation relied on Softr and separate spreadsheets. Matches were coordinated manually, data was scattered, and compatibility scoring, reminders, feedback loops and hard business-rule enforcement were missing.
The platform handles sensitive dating and personality data under GDPR. The rebuild had to replace a live business system without losing its customer base, while enforcing rules such as no repeated matches, identity verification, one active match per person and contractual date guarantees.
The approach.
I chose a full rebuild rather than adding more workarounds to the no-code system. The platform was developed full-stack by one person over approximately six months and split into thirteen domain modules covering authentication, profiles, matching, planning, contracts, communication, feedback, intake and administration.
Critical logic was written test-first. A penetration test and remediation preceded launch, followed by a GDPR-filtered migration of more than 2,000 existing users. Modelling the rules and data first, then testing the risky paths, then migrating, kept the live transition controlled.
Architecture and decisions.
The architecture is serverless-first to minimise operational work, with AI used as decision support under human control.
- ·
SvelteKit on Cloudflare Workers. The platform runs at the edge, scales automatically and needs no dedicated operations team.
- ·
Neon Postgres with Drizzle. Matchmaking is relational: 39 tables use foreign keys, unique constraints, transactions and indexes to enforce integrity.
- ·
Better Auth, self-hosted. Roles, sessions and TOTP two-factor authentication remain under direct control without per-user identity SaaS pricing.
- ·
Google Gemini through REST. The model is called directly in a way that fits the Workers runtime and always remains subordinate to deterministic filters and human approval.
- ·
Stripe, Twilio, Resend, R2 and KV. Payments, SMS and WhatsApp, email, European photo storage, caching and rate limits complete the platform.
The hardest problems, solved.
- 1
Reliable, GDPR-aware AI matching. All candidates are scored deterministically first; only the top ten reach the model. Profile text is stripped of names and personal identifiers and used only with consent. Matchmakers approve every proposal, and deterministic scores remain available if the model fails.
- 2
Serverless database traffic. A client per request plus Hyperdrive pooling avoids cross-request socket reuse, while waitUntil keeps emails and messages alive after the response completes.
- 3
Business rules under concurrency. Transactions enforce one active match per person, scheduling slots are locked safely and a normalised unique pair index prevents repeated matches.
The result.
Matched By Us runs in production and fully replaces the previous Softr system. Intake, matching, planning, dates, feedback and contracts now form one workflow, and more than 2,000 users were migrated with GDPR filtering. A suite of 4,668 tests protects ongoing changes.
The same approach applies when a company has outgrown spreadsheets, Airtable or no-code tooling: formalise the data and rules, automate communication, add carefully supervised AI where it improves decisions, and deploy on infrastructure that does not require a separate operations team.
Under the hood.
Frequently asked questions.
Is your question not listed? Email me directly.
Can this approach support our planning, intake or billing processes? +
Yes. The core is converting manual, rule-based work into a relational model, enforced rules and automated communication. Matching is the domain here, but the architecture applies to planning, intake, quotations, billing and document processing.
Can AI analysis work for our use case? +
Often, but the value lies in how it is bounded. Deterministic rules handle what must be certain, AI contributes where interpretation helps, and a person retains the final decision.
How long did the platform take? +
The complete platform was built by one developer in approximately six months. A focused first release for another process can often be delivered in weeks, depending on scope.
What would a platform like this cost? +
The first release needs to be defined before a credible fixed price can be given. This project demonstrates that one developer can deliver and operate a complete business platform; your proposal would phase the essential workflows first.
Is sensitive data protected? +
The design includes role-based access, TOTP for staff, consent checks for AI, removal of personal identifiers from prompts and logs, European photo storage and a GDPR erasure workflow. The platform was penetration-tested before launch.