I quit my 6-figure SRE job at Bank of America 2 weeks ago to build AI engineering interview prep. Here's where things stand.
THE PROBLEM I KEPT SEEING
Every engineer I knew preparing for AI roles at Anthropic and OpenAI was failing their technical rounds. Not because they weren't good enough, but because nothing prepared them for what these companies actually ask. AI engineering interviews in 2026 test RAG pipeline architecture, token efficiency, agent orchestration, and prompt injection defence. LeetCode covers none of this. Nobody was built specifically for AI engineering. So I quit and built it.
WHAT I BUILT
Velocode (velocode.ai) — interview prep for AI engineering roles. Three independent AI agents score every submission simultaneously on token efficiency, architecture, security, and correctness. Then Claude and GPT-4o both attempt the same problem, cross-score each other, and we synthesise the golden answer from both. No single model produces this alone. The interview simulator reads your actual submission history and builds a session targeting your lowest-scoring dimensions. It knows what you're bad at before the session starts.
WHERE THINGS STAND (honest)
These are early numbers — the platform launched yesterday.
Subscribers: 24. Paid users: 1. MRR: $20. Problems published: 126. Simulator sessions: 7.
WHAT I GOT WRONG ALREADY
The onboarding was showing a paywall before users had seen any value. 24 signups and almost nobody made it to the practice page. Fixed it today — new flow sends users directly to their first problem with zero friction. Also, welcome emails were going to Promotions. Spent this morning rewriting all 4 as plain text from "Sanjna" instead of "Velocode AI."
WHAT'S WORKING
The founding story post on LinkedIn drove most signups. The people who make it to the score reveal page stay — the scoring is genuinely differentiated, and users feel it.
WHAT I'M BUILDING NEXT
SEO guides for "Anthropic interview questions" and "RAG interview questions," university outreach offering free Pro access to AI/ML clubs, and a B2B assessment platform in month 2.
Happy to answer questions about the technical architecture, scoring pipeline, onboarding mistakes, or what it's like to quit a stable job with no safety net. Building this fully in public. Follow along if you want the real numbers every week.
Always open to feedback, tips, or just to chat about the journey.: sanjna@velocode.ai
Top comments (0)