Sector · Gig economy
Marketplaces that match, pay and keep workers moving.
We have designed, built and operate the full stack behind on-demand work — from the first WhatsApp message to the last penny of a payout. Two-sided, high-stakes, and unforgiving of downtime, because every outage is someone's income.
What we build
Every part of a two-sided marketplace, from a team that has shipped one.
Matching that explains itself
A multi-stage pipeline: hard eligibility filters first, then full-text and vector search over worker profiles, then a ten-factor weighted score — skill relevance, availability, responsiveness, experience, proximity, qualifications, equipment, completion rate and more — with an LLM semantic pass on top. Every score is decomposable, so a human can see exactly why a worker ranked where they did.
Dispatch where workers actually are
Offers go out over WhatsApp with accept and decline buttons, backed by SMS, push and email. Each tap carries a signed action token, so a worker can accept a shift from a lock screen without opening an app. Offers expire on a fixed window and the next candidate is lined up automatically.
Onboarding and right-to-work built in
Conversational, AI-assisted onboarding gets a worker from first contact to a complete profile in minutes. Right-to-work runs through Home Office share codes and passport checks with face match, permission expiry and work-restriction tracking — and the platform never offers work to anyone who is not verified.
Payments that hold, capture and reconcile
Stripe Connect underneath: funds are authorised when a worker accepts and captured on completion, so buyers are protected and workers are certain of being paid. Platform fee, VAT, processing fee and net-to-worker are stored per transaction against the payment, charge, transfer and payout identifiers — a reconciliation spine, not an afterthought.
Ratings, disputes and no-shows, handled
A sixteen-state gig lifecycle covers acceptance, progress, dual-sided completion, cancellation by either party, disputes and no-shows. Reviews carry would-work-again and per-skill attribution and feed back into matching. If a buyer goes silent after a shift, completion is auto-confirmed and the worker is still paid.
Enterprise buyers, not just individuals
Organisation accounts with domain-based invites and admin and booker roles; mass-hire requests that fill multiple roles from a single brief; consolidated invoicing restricted to organisation members; calendar sync and a full admin portal with audit, analytics and a human-in-the-loop review queue.
Field notes
The scar tissue that comes with the playbook.
Marketplaces fail in the details. These are a few of the details we have already paid for.
The matching engine learns from its reviewers
Every human approve-or-remove decision in the review queue is mined on a schedule to recalibrate score thresholds and factor weights. The system gets better at pre-empting its reviewers without anyone retraining a model by hand. On a production regression across sixty distinct skills, it produced zero wrong-job matches.
We stopped trusting the vendor's verdict
A right-to-work provider began wrongly rejecting workers with EU Settled Status — eleven of twenty-seven checks in a month. Rather than accept the verdict, we now classify directly from the authoritative gov.uk conditions text and route anything uncertain to manual review. Workers stopped being excluded for a vendor's bug.
Pricing is snapshotted, not recalculated
The terms and price a buyer accepts are frozen against the gig at the moment of acceptance and protected by golden-fixture regression tests. Fee changes never retroactively alter what someone agreed to pay.
Notifications have an SLA
Four channels — WhatsApp, SMS, push and email — with per-user preferences and delivery logging. Critical events target sub-minute delivery at better than ninety-five percent success, because a missed shift offer is lost income.
What comes with it
The platform patterns, integrations and operating discipline we bring to a new marketplace on day one.
- Next.js, React and TypeScript, server-actions-only architecture
- PostgreSQL with pgvector for semantic worker search
- Stripe Connect with manual-capture holds and connected worker accounts
- WhatsApp Cloud API, Twilio SMS, push and email as first-class channels
- LLM agents for onboarding, brief extraction, matching and incident detection
- Companies House and Home Office integrations for buyer and worker verification
- GDPR by design: retention clocks, deletion archives and PII redaction crons
- Available under our fixed price model, then run under managed service