Interview-prep tools fall into two camps: static question banks that never change regardless of the role, or subscription courses that lock users into recurring payments for content they may use once. Neither reflects how people actually prepare — in short, focused bursts before a specific interview, for a specific role, at a specific company. The First Interview set out to build an AI-native alternative: personalised, AI-generated mock sessions with real answer evaluation.
Rorix partnered on the complete foundation: a pay-per-use credit economy priced against real AI token costs, a dual-model AI pipeline with automatic failover, an institutional tier for colleges and placement cells, growth mechanics without dark patterns, and the marketing-site architecture — every pricing tier margin-verified across 60+ session-type combinations before a single price was published.
Static question banks ignore the role, the company, and the candidate; subscription courses charge monthly for naturally bursty usage. The product had to be personalised to an actual job description or resume, and monetize per use.
Delivering questions is easy; the value is feedback. Every session had to evaluate the candidate’s submitted answers against a benchmark and produce a full performance report, reliably, at a predictable cost.
Individual job seekers buy one session before an interview; colleges and placement cells run mock-interview drives for hundreds of students at once. Both had to be first-class from day one, not a retrofit.
Session cost depends on generation type, difficulty, question count, and optional behavioral rounds. Flat pricing would either bleed margin on heavy sessions or overcharge light ones — and a provider outage could take the product down.
Designed a credit economy where session cost is calculated from generation type (technology / job description / resume / resume+JD), difficulty, question count, and behavioral rounds — so pricing scales precisely with AI compute cost. Credits never expire.
A two-call pipeline: the first call generates personalised questions with internal “ideal answers”; the second evaluates the user’s submissions against that benchmark into a full report. Gemini Pro 1.5 primary for cost efficiency, Claude Sonnet 4 as automatic fallback behind a circuit breaker.
A separate bulk-purchase tier for colleges and training institutes: shared credit pools, per-student usage caps, an admin dashboard, and GST-compliant invoicing — priced to make bulk mock-interview drives significantly cheaper per student.
Every credit pack and pricing tier was modeled and verified against real AI token costs across 60+ session-type combinations to guarantee consistent gross margins before any price went live.
A first-purchase bonus window, time-boxed signup credits, referral rewards, and a streak-and-badge system drive early conversion and retention without subscription lock-in.
Scoped a Next.js marketing and lead-generation site with a token-based themable design system and programmatic SEO pages per technology and role, over NestJS services and PostgreSQL — with the database schema, security model, refund policy, and API architecture fully documented.
Margins, failover, and pricing are architecture decisions. Let's design your AI product so the unit economics work before launch.
Everything on this topic that lives elsewhere on the site.