AI candidate sourcing for recruiters

The problem

A US recruiting firm had a talent intelligence platform built as a Lovable prototype. It could run a search in a demo. It couldn’t bill customers, handle real client data or hold up under real searches.

What we did

We built the backend on Supabase. Edge functions run candidate search and enrichment through third-party data APIs, and role classification logic sorts job titles so searches return the right people. Stripe handles subscriptions and credit packs. One table owns every account’s tier, so billing, usage limits and feature access can never disagree with each other.

We also built the admin tooling the team runs the platform with, and led the QA rounds before launch, including pressure tests against the live data provider.

One lesson shaped the whole build. Several edge functions existed only in the deployed Supabase project, not in the Lovable repo. So every diagnosis ran against what was actually deployed, never against the local code. That rule kept false alarms away from the client.

The result

A stable production app the team uses every day with real client data. In the client’s own review, it was probably the most complex project we had taken on.