A US payments-infrastructure startup, post-Series A, needed a product engineer who could own banking-API integrations end-to-end. Their roadmap included eight separate integrations over six months — Plaid for bank account verification, Stripe Treasury for stored balance, Modern Treasury for ACH routing, plus five smaller niche providers. They wanted someone who'd already shipped integrations with at least three of these in production, not someone who would learn on their dime. The role was product-engineer in the truest sense: spec, design, build, ship, and own ongoing maintenance per integration.
Banking-API integration is one of those skills where the production scars matter more than the codebase familiarity. The APIs themselves (Plaid, Stripe Treasury, Modern Treasury) are well-documented; the challenge is the edge cases that only surface under production load. Account-verification webhooks that arrive out of order. ACH return codes that mean different things at different banks. Reconciliation when the bank's settlement file disagrees with your transaction log. An engineer can pass a generic integration interview but ship code that gets reconciliation wrong by 0.005% — which on a payments-infrastructure product is a regulatory event. Most engineers who claim banking-API experience on their resume have built a Plaid Quickstart and shipped to a sandbox — they've never had the 3am Slack message about a stuck ACH return at month-end. The client had been searching for 8 weeks via internal recruiting and one external agency, with two offer rejections and one ghosted offer. Their CTO was getting frustrated; the eight-integration roadmap was already slipping.
The AI scoring rubric was the most aggressive we've ever shipped: any candidate whose resume mentioned banking integrations got cross-referenced against verifiable employer metadata. If they claimed Plaid experience but worked at a firm without verifiable Plaid in their stack (we maintain a public-stack registry), they were downweighted to near-zero. That filter alone took 1,178 to 412 — most banking-API-mentioning resumes failed the verification step. Senior recruiter screen cut to 88, focused on the 'has been on the 3am stuck-ACH Slack thread' signal that emerges from reference calls more than resumes. 30 of the 88 completed a 4-hour written assignment: design a Plaid-to-Stripe-Treasury integration with proper retry semantics on the bank-account-verification step, including handling for out-of-order webhooks and partial-success scenarios. 14 cleared (most failed on the webhook ordering — they assumed in-order delivery, which is wrong). Live coding (90-minute) cut to 8. The founders ran 90-minute product-engineer final-rounds with the top 6 — explicitly testing judgment calls (what do you ship vs defer? where do you cut scope when the roadmap slips?). The winning candidate (Hyderabad-based, 7 years experience, previously at a Singapore-based payments-infrastructure firm) had shipped eight banking integrations himself and walked through the founders' roadmap with specific concerns flagged on three of the eight — exactly the product-engineer judgment they wanted.
Offer day 11 at top-of-band. Accepted within 36 hours. Started day 16 with onboarding compressed by pre-provisioning during offer acceptance. The engineer shipped his first integration (the Plaid bank-account-verification flow) to production in week three of the engagement. Over the next 12 weeks he shipped 7 more integrations on schedule — the eight-integration roadmap originally scoped for six months landed in four. Per-integration time fell from a baseline 5 weeks to 1.5 weeks as he established a reusable integration framework (retry semantics, webhook handling, reconciliation patterns) that the rest of the team could use for future integrations. Reconciliation accuracy on production transactions: 99.998% (one mis-reconciled transaction per ~50,000), well within the regulatory tolerance. The engineer is still with the team 12 months in; converted to full-time in month seven with a small equity grant. Client returned for two more roles (a senior data engineer and a Staff frontend) in the following six months.
Two-hour call with both founders. Mapped the eight banking-API integrations they had on the roadmap, their order, and the production lessons from the two integrations they'd already shipped (so we could test for similar patterns).
AI scoring downweighted any candidate who hadn't verifiably shipped a banking-API integration in production. The 1,178 applicants became 412 after this filter — most banking-domain candidates work in payments-firm full-time roles, not as freelancers.
30 candidates completed a 4-hour assignment: design a Plaid-to-Stripe-Treasury integration with proper retry semantics on the bank-account-verification step. 14 cleared. Top 8 went to live coding with our senior recruiter.
Founders ran 90-minute product-engineer interviews with the top 6 — focused on judgment calls, not algorithms. Offer day 11. Started day 16.
For niche domains where production scars matter more than framework familiarity, employer-metadata verification is more reliable than self-reported resume claims. We've built a public-stack registry of firms with verified production usage of major banking APIs — cross-referencing candidate employment against that registry catches 90% of the resume-inflation cases. The second lesson: product-engineer hires need a product-engineer final round, not a coding final round. Asking 'where would you cut scope?' surfaces the judgment that distinguishes shipping engineers from coders — and that judgment is what makes the difference on a roadmap with eight integrations and limited time.