Building from 0 to 10k Users: What I Learned as a Founding Engineer at an AI Startup

Joining Kamyab as a founding engineering member was one of the most intense, high-velocity learning experiences of my career.

We set out to build an AI-powered NEET preparation platform that helped competitive exam students practice with adaptive question engines, instant explanations, and analytics.

Within two months of launch, we scaled to 10,000+ registered student signups and facilitated over 500,000 question attempts. Here is what that journey taught me about engineering speed, product intuition, and leveraging modern AI workflows.

#1. AI-Assisted Workflows as a 10x Force Multiplier

Before Kamyab, my primary expertise was heavily frontend: React, TypeScript, state management, and modern CSS architectures.

When you are at seed stage with limited headcount, you cannot afford to say "that's not my stack". The platform required high-throughput backend services for quiz generation and scoring. We chose Fastify on Node.js for its low overhead and robust schema validation.

I leaned aggressively into AI-assisted development tools (Cursor and Claude Code). Instead of spending weeks reading through documentation and boilerplates, I paired with AI to:

• Generate typed schema definitions with TypeBox and Fastify.
• Draft unit and integration tests for exam evaluation algorithms.
• Refactor complex database queries and indexing strategies.

This enabled me to ship production-grade backend endpoints within days rather than weeks, keeping our sprint cycles extraordinarily tight.

#2. Speed of Shipping vs. Architectural Cleanliness

One of the hardest psychological shifts for an engineer moving into a startup is accepting that shipping the wrong thing slowly is far worse than shipping the right thing quickly and refactoring later.

In the beginning, we made two critical rules:

1. Strong TypeScript types everywhere: No loose any. This prevented 90% of regressions while moving fast.
2. Modular directory boundaries: Keep business logic out of UI view components so that when business requirements shifted, we only touched domain modules.

When a feature proved successful with users, we spent scheduled technical debt cycles refining database indexes, adding Redis caches, and hardening error boundaries.

#3. The Power of Distribution Loops

An engineering marvel that nobody knows about is worthless. At Kamyab, engineering worked hand-in-hand with growth:

• We built viral report cards and performance analytics that students eagerly shared on WhatsApp groups and Telegram channels.
• We partnered with educators and student mentors to iterate on features they directly asked for.
• Fast turnaround on bugs reported by students built immense goodwill. When a student encountered a rendering glitch on an obscure Android browser and we pushed a fix within 2 hours, they became our biggest champion.

#4. Key Takeaways for Early-Stage Engineers

1. Be product-oriented, not just code-oriented: Understand the user's pain point before debating library choices.
2. Embrace AI pair programming deeply: Treat modern LLM tooling as an intelligent junior engineer that writes boilerplate, finds edge cases, and accelerates prototyping.
3. Instrumentation from Day One: Add basic error logging (Sentry) and user session telemetry early. You cannot fix what you cannot see.