Building Maxpeak
How a 3-stage AI-powered vetting pipeline solved the Egypt-to-Saudi talent trust problem — connecting elite Egyptian engineers with Saudi companies through verified, guaranteed hiring.
01The Problem
TheEgypt-Sauditalentcorridorhasatrustproblem.
Saudi companies need software engineers. Egypt has a large pool of talented engineers hungry for premium opportunities. The corridor exists — but no trusted platform serves it. Global platforms like Toptal and Andela are expensive for SMBs and lack regional focus. Local job boards are full of noise — unverified profiles with inflated CVs and no quality signal.
The root issue is trust asymmetry. Saudi companies can't reliably evaluate Egyptian candidates remotely. Egyptian engineers can't prove their quality without a trusted intermediary. Every hire is a leap of faith — and when it fails, both sides lose time and money with no recourse.
Maxpeak's answer was a rigorous 3-stage vetting process that makes the verification work happen before the hire — so by the time a company sees a profile, the quality question is already answered. And to back it up: a free replacement guarantee if the hire doesn't meet expectations.
Core requirements
- 3-stage vetting pipeline: AI screening, expert interview, profile verification
- Verified badge system with quality guarantee for companies
- Contact protection — companies and engineers can't bypass the platform to connect directly
- Bilingual (Arabic/English) from day one — both markets need it
- In-app chat, analytics dashboards, and integrated payments for subscriptions
02The Vetting Pipeline
Trustisbuiltbeforethefirstinterview.
The entire product thesis rests on this pipeline. Every engineer in the talent pool has passed all three stages — companies browse only verified talent.
Stage 1 — AI Screening
Automated profile analysis, skill extraction, and initial scoring
Stage 2 — Expert Interview
Senior engineer assessment — live technical evaluation
Stage 3 — Profile Verification
Portfolio review, reference checks, background validation
Verified Badge + Quality Guarantee
Approved engineers get a Verified badge. Companies get a free replacement guarantee if the hire doesn't meet expectations.
03Key Decisions
Whatwechoseandwhy.
AI-first screening to scale the top of the funnel
Manual review of every applicant profile doesn't scale. We built an AI screening layer that extracts skills from CVs, scores candidates against a rubric, and surfaces the top candidates for expert review. This cut the expert review workload by ~60% while improving signal quality — reviewers spend time on candidates who already passed an objective bar.
Contact protection by design, not policy
Marketplaces bleed when users go around the platform. We built contact protection at the data layer: engineer contact details are never returned by the API. In-app messaging is the only channel, and it's logged. This keeps all relationships platform-mediated and enforceable by the quality guarantee.
Bilingual from the first line of code
Adding i18n to an existing app is painful. We used Django's i18n framework from day one — all user-facing strings are translation-wrapped, all dates and numbers are locale-aware. The Arabic UI required full RTL layout support in React 18, handled via CSS logical properties rather than mirrored stylesheets.
Quality guarantee as a product feature, not a promise
A replacement guarantee only works if it's enforceable. We built a structured dispute flow into the platform: companies can open a replacement request within a defined window, which triggers a review process and, if valid, re-enters them into the matching pool at no cost. The guarantee is code, not just copy.
04Outcomes
Amarketplacebuiltonverifiedtrust.
Stage vetting pipeline
AI screening → expert interview → profile verification
Markets served
Egypt (supply) · Saudi Arabia (demand)
Replacement guarantee
Free replacement if the hire doesn't meet expectations
Languages supported
Arabic and English, bilingual from day one
05Challenges & Lessons
Wherethehardproblemslived.
AI screening calibration
The AI scorer needed to be calibrated against real expert review outcomes. We ran a shadow mode for the first 200 applications — AI scores were logged but not shown — then tuned the scoring weights against expert decisions.
Chat at scale
In-app messaging sounds simple. Enforcing contact protection without making the UX feel like a walled garden required careful message filtering, rate limiting, and a clear escalation path for disputes — all without false positives that block legitimate communication.
CV parsing variance
Egyptian engineers submit CVs in wildly different formats — Word, PDF, image scans, mixed Arabic/English. Building an extraction pipeline that reliably pulls skills and experience from unstructured documents required iterating through several NLP approaches before settling on a hybrid rule + model approach.