Own platform

Recruit Portal: a live, self-healing hiring tool

A live hiring-evaluation tool that streams matched results from seven connectors, with a watchdog that restarts on three failed health checks and rolls back on six, unattended.

Era 6, the AI-agent-native practice · Live

The situation

EMC Notary, our own remote-notary business, needed a way to evaluate hiring candidates against live job-market data pulled from multiple outside sources at once. Building that by hand, one data source at a time, would have taken weeks and left us with a tool nobody trusted enough to hand to an outside evaluator. So we built it as a governed product from the start, one an outside evaluator could be handed without a caveat.

What we built

Recruit Portal is a hiring-evaluation tool, live at recruit.emcnotary.com, that pulls candidate and job data from seven outside connectors (job boards, applicant-tracking systems, and enrichment services) and streams matched results to the screen as they arrive, instead of making the user wait for one big report. It runs on a Raspberry Pi (a small physical computer we host ourselves) reached through a Cloudflare tunnel (a secure, no-open-ports connection from our server to the public internet).

Named approval gates

Every release goes through a canary step (a term for testing a new version on a small, isolated slice of traffic before it reaches real users) before it is exposed to an evaluator. A watchdog (an automated monitor) checks the live site’s health once a minute. Three failed checks in a row restart the service on their own. Six failed checks in a row roll the release back to the previous working version automatically, no human needed to notice or intervene in the moment. A human still decides when a new version goes live in the first place and who receives access.

Verified results

  • 172/172workspace test suite passingrecruit portal workspace
  • 3 failed checkstriggers an automatic restartrecruit-eval watchdog timer
  • 6 failed checkstriggers an automatic rollbackrecruit-eval watchdog timer

The full workspace test suite passes at 172 of 172. The automated health check and rollback system is armed and verified on the live host. The deployed build was checked line by line to confirm no internal scoring logic or data-source details ever reach the browser, only the final matched results.

What this means for your business

This is the same pattern we would bring to your hiring or operations tooling: a real, live system with its own safety net, so a bad release gets caught and reversed automatically before anyone notices. You get a working tool and a record of how it stays reliable while unattended.

Talk about a build like this

Back to the full build history · Get in touch