Talentrah: an AI career platform built for the market it serves

Applying for a job properly means rewriting your CV for every role. Read the posting, work out what it is really asking for, rephrase your experience to match, write a covering letter, keep track of where you sent it. It takes hours, it is easy to do badly, and most people either skip it and get filtered out, or burn an evening per application.
The tools that solve this exist. They are built for the United States: American ATS integrations, American salary data, American CV conventions, and pricing that starts around $24 a month. The cheapest of them costs more than a tenth of the average monthly income in Nigeria. For a job seeker in Lagos or Abuja, the product that would help most is the one they cannot justify buying.
Three constraints shaped almost every decision, and none of them appear in a US-first product spec. Devices and data cost money: the audience skews toward mid-range Android phones and metered mobile data, so payload weight was treated as a product requirement, not a performance nicety. Mobile money is a push, not a pull: card subscriptions renew silently, but wallet-initiated payments require the user to approve each charge, so a subscription model that works perfectly on cards quietly fails on the rail most users actually have. And trust is scarce and easily spent: Nigerian job seekers are heavily targeted by recruitment scams, so a platform that plays fast and loose burns its credibility before it earns any.

Talentrah is an AI career platform where the intelligence is genuinely useful and the economics work at Nigerian prices.
What shipped: a free, no-account demo on the landing page, where pasting a job description returns a match score, a gap analysis and a tailored CV preview without signing up. Match scoring with an explanation, so Farah, the platform's AI copilot, can answer “why is this 82%?” rather than leaving the number unexplained. JD import and CV tailoring that parses a posting into structured requirements and returns specific, checkable feedback, plus an editable tailored CV and covering letter. Auto-apply with review before submission, and a log of what was sent where. A job feed built from an aggregation pipeline, deduplicated and freshness-checked as background work so it never blocks the interface, with externally sourced roles labelled as such. A CV builder, application tracker, scholarship discovery and referral programme. Self-serve employer posting with company verification by work-email domain. And a credits model rather than a subscription, so someone in an active job search can spend for a fortnight and stop.
Four decisions are worth explaining. Auto-apply asks before it sends: the obvious version submits applications while the user sleeps and the volume metric goes up, but it also trains recruiters to treat the platform’s applicants as spam, so Talentrah requires review before submission. Payment rails were a day-one architecture decision, because the wrong choice there is not a refactor but a rebuild of everything that touches billing. We refused to scrape what we should not: job aggregation runs on official APIs and structured data, and scholarship listings pass a human review gate before publishing, because a stale job posting is an annoyance but a wrong scholarship deadline costs someone a year. And nothing on the product claims what it has not earned: no invented user counts, no borrowed trust badges, and an FAQ that says plainly which features are live and which are planned.
How it is built: a TypeScript monolith, Next.js 16 on Vercel with React Server Components and Server Actions, Supabase Postgres for data and auth, Paystack for payments, and Google Gemini with Groq as a switchable second provider behind a single interface. No microservices, no bespoke infrastructure. Security is enforced at the database rather than the application, so row-level security means a bug in application code cannot leak another user’s data. The schema is versioned and auditable, with ninety migrations in the repository and generated types so a schema change breaks the build rather than production. There are 1,136 automated tests across 109 files, plus end-to-end coverage, and CI runs a secret scan before it runs anything else.

Talentrah launched on 1 September 2026 and is open to sign-ups. It is too early for usage results, and this page will not pretend otherwise: the numbers go here when they exist, not before.
What can be said now is what was delivered. The scope above is live, and the marketing site at talentrah.com is ours too. What is deliberately not there yet is stated on the product itself: human mentorship is designed but not launched, and skills verification, the employer talent directory and the full self-serve advertising platform are specified and sequenced behind evidence of real usage rather than shipped on optimism.
Zimcrest carried every decision in this product: the market research, the pricing model, the architecture, the design, the build, the test suite, and the choice of what to leave out.