AI automation that cut recruitment admin by 80%.

The challenge

The HR team was entering every candidate by hand across three tools that did not share data. Job descriptions lived in one place, applicants in another, and the matching happened in someone's head. Roles stayed open longer than they should have, and good candidates aged out of the pipeline while the team was still doing admin.

The cost of that fragmentation was not just time. Every hour a role stayed open carried a compounding cost: lost productivity, slower revenue, and increased pressure on teams that were already stretched. The client estimated that a single senior hire delayed by two weeks cost more in lost output than a month of recruiting fees. The manual process also introduced bias, because whoever reviewed CVs last made the call, with no systematic baseline to compare against.

Good candidates were slipping through entirely. Because the pipeline had no scoring or ranking layer, applicants who applied early and matched well sat in the same queue as poor fits who applied later. There was no mechanism to resurface strong candidates once a role moved forward, which meant the same sourcing work had to happen again each time a similar position opened. Institutional knowledge about what a good hire looked like for a specific role was never captured anywhere.

What we built

We built Sidechain, an AI-powered recruitment platform. Company and role data is pulled automatically from Crunchbase. Candidates sync in real time from Folk CRM via Zapier. An AI matching engine, built on Claude Sonnet and Gemini, scores every candidate against every open role and surfaces ranked shortlists in a fast React and Supabase dashboard. No manual entry. No context switching.

The decision to combine Claude Sonnet and Gemini was deliberate. Claude Sonnet handles the nuanced language tasks, reading role briefs, interpreting candidate summaries, and producing a structured rationale for each score. Gemini handles broader factual lookups and cross-referencing, including company context from Crunchbase and signals about the hiring environment in a given sector. Using both models in sequence gave the scoring engine more coverage than either model alone could reliably produce.

Supabase was chosen for the dashboard because it gave the client a real-time data layer without the overhead of managing a separate backend. Every scoring event, candidate update, and role change writes immediately to the database and surfaces in the recruiter view without a page refresh. The React frontend is deliberately minimal: ranked shortlists, score breakdowns, a one-click handoff to interview scheduling. Nothing the recruiter does not need.

The scoring engine evaluates each candidate across four dimensions: skills alignment against the role brief, seniority calibration based on tenure and title progression, cultural signals drawn from public writing and prior company contexts, and availability indicators that flag whether outreach is likely to land. Each dimension is weighted per role type, so a technical hire and a commercial hire use different scoring distributions. Recruiters can see the full rationale behind each score, not just the number.

Day to day, the recruiter experience is fundamentally different. A new role goes in as a brief, Crunchbase and Folk populate automatically, and within minutes the scoring engine returns a ranked shortlist with explanations. The recruiter reviews fit, moves candidates forward with one action, and the CRM updates without any manual input. What previously took two to three days of sourcing and sorting now takes under an hour for a well-defined role.

The outcome

Candidate matching became 50% faster from brief to shortlist. Hiring accuracy, measured against 90-day retention, improved 30%. Manual data entry fell 80%. The team now spends time on conversations, not spreadsheets, and the platform scales to handle 10x the volume without adding headcount.