← all projects
Case study · full-stack GenAI product

PsyMetric AI
a GenAI product with real users — and real payments.

An adaptive career-discovery platform: psychometric assessment powered by Gemini, profiles mapped against 140+ industry career benchmarks, personalized 4-year roadmaps, automated PDF reports — and a working Razorpay paywall. Not a demo; a product people paid to use.

Repository ↗
89%
USER SATISFACTION
<200ms
API RESPONSE
200+
CONCURRENT USERS
140+
CAREER BENCHMARKS
PsyMetric AI landing page
The landing page — React + Vite + TypeScript frontend.

01 · The product

Career guidance for students is usually generic. PsyMetric makes it adaptive: the backend evaluates each response with Gemini (via Vertex AI) and tailors subsequent questions to the user's background — cognitive traits, interest fields, behavioral style — then maps the resulting profile against 140+ career benchmarks to generate matches and a personalized 4-year academic roadmap.

Because it's a product and not a notebook, the unglamorous parts got engineered too: quiz autosave & restore (progress persists in PostgreSQL — close the tab, resume later), one-click restore of previously unlocked reports after retakes, and automated PDF report generation.

Adaptive assessment workflow screen
Adaptive assessment — questions adjust to prior answers via Gemini.
AI career match results screen
Career matches — profile scored against 140+ benchmarks.

02 · The engineering

FastAPI backend with 12+ RESTful endpoints holding <200ms response times, scaled to 200+ concurrent users; React + TypeScript SPA frontend; PostgreSQL for sessions, profiles and reports; Gemini integration with inference optimization for real-time personalized mapping. Deployed and operated in production — 50+ profiles processed, 15+ roadmaps generated, 89% satisfaction.

And the part most student projects never touch: payments. Razorpay checkout with a coin-based credit system (₹299 per assessment) gating report access. Integrating a payment gateway means handling real money, real failure modes, and real user trust.

Razorpay payment integration screen
Razorpay checkout — coin-based credits, ₹299 per test.
Personalized career roadmap screen
The deliverable — a personalized, chronological 4-year roadmap, exportable as PDF.

03 · Why it matters

Products teach what models can't.

Sessions, payments, autosave, edge cases, angry users — the distance between "the model responds" and "someone pays and gets value" is where the engineering lives.

LLMs need scaffolding, not just prompts.

Adaptive questioning, benchmark mapping, and report generation are structured pipelines around Gemini — the model is one component, not the system.

It compounds.

This project led directly to my current internship building RAG-based assessment pipelines at Suresh Dani Classes — same domain, production scale.