Recruitment

We turned a pile of unstructured CVs into a ranked, searchable talent engine — sourcing, screening and interviewing candidates on autopilot, with humans in control.

HR Tech / Recruitment (MENA — Egypt, UAE, KSA)

Book a call Visit Skatch

skatch.com

Ranked matches

5 of 1,240 scored

1
Candidate A94
2
Candidate B88
3
Candidate C81
4
Candidate D72
5
Candidate E64

The challenge

Skatch's two-sided hiring marketplace (800+ employers across Egypt, UAE and KSA) was drowning in unstructured CVs and slow, manual screening, while candidates dropped off before finishing applications. Recruiters couldn't find relevant people fast, and an early match-score engine produced unreliable rankings that hid the best candidates.

Recruitment

HR Tech / Recruitment (MENA — Egypt, UAE, KSA)

What we built

The system, in parts.

CV intelligence engine that parses unstructured PDF/DOCX resumes (bilingual English/Arabic) into structured candidate profiles — skills, work history, education — with additive backfill that enriches rather than overwrites existing data

JD-vs-CV match-score system scoring candidates on skills, experience recency, industry relevance, education hierarchy and location, using per-job GPT-generated scoring weights and vector embeddings to rank applicants

Natural-language CV Finder for recruiters: free-text queries ("sous chef in Dubai, 3+ years") parsed into structured filters with seniority-aware, bilingual job-title expansion so search respects the role tier

WhatsApp AI agents that build candidate CVs conversationally and let people apply to jobs directly in chat — plus a Mass Hiring WhatsApp engine for multi-channel candidate sourcing and event outreach at scale

AI onboarding assistant and personalized job-recommendation + job-match email workflows that re-engage candidates and route them to the most relevant roles

AI voice-interview system that conducts structured STAR/competency-based interviews and produces consistent, scored evaluations

Outcomes

What changed for them.

Manual CV screening collapsed from a reading exercise into structured, searchable profiles — recruiters find relevant people in minutes, not hours
Match scores became trustworthy: eliminated false zeros for partially-relevant candidates and impossible >100 / NaN scores, so the genuinely best applicants now surface at the top
Candidate sourcing and applications run where people already are — WhatsApp — lowering drop-off and opening multi-channel, bilingual reach across MENA
Re-engineered to be cost-efficient: scoring moved to lighter models and cheaper embeddings, cutting cost per application roughly 20x with no loss in ranking quality
Reliable by design — validation suites and head-to-head candidate sampling confirmed correct rankings before rollout, with all existing API contracts preserved (zero downstream breakage)

How it’s built

The stack.

Node.js / ExpressMongoDB / MongooseOpenAI GPT-4o & GPT-4o-minitext-embedding-3-small (vector matching)WhatsApp Business / Meta Graph APINext.js (HR portal)AWS S3 / CloudFrontSocket.IOpdf-parse / Mammoth (document extraction)

More work

See all

Healthcare

Automated compliance for an enterprise health plan — provider-directory accuracy, CMS network adequacy, and audit-readiness run as an always-on program instead of a manual fire drill.

Read the story

Government

A Punjab government counselling engine that ranks and allots seats across 200+ colleges and multiple rounds — automatically, transparently, and provably at scale.

Read the story

Marketing Ops

We turned a 50-year, 400+ product irrigation manufacturer's catalog into an always-on marketing engine — programmatic SEO, campaign content, and lead capture at global scale, without a big in-house marketing team.

Read the story

Two ways to start

Want a system like this?

Tell us the workflow you want to run itself. We will scope a focused first project — designed, built, and operated, with humans in control.

Book a call

See our work