speakeasyprep.ai
A working AI interview prep tool shipped to real users on a real stack. After the no-code prototype proved there was a there there.
Generic interview prep tools give you a question bank and a timer and call it a day. But the job of interview prep isn't reading questions. It's rehearsing your answers against a specific job description, hearing yourself out loud, and knowing which stories land and which ones ramble. None of the existing tools did that end-to-end. Candidates were doing the work in Google Docs, Loom, and their heads.
Ship the minimum loop on no-code first (Lovable plus n8n) to prove the workflow was worth building. The moment it was validated, throw out the prototype and rebuild the whole thing on a stack that could handle real users, real payments, and real AI latency: Next.js 16 on Vercel, Supabase for auth and data, and a multi-model AI pipeline chosen per task instead of one model for everything. The rebuild wasn't scope creep. It was the plan.
- Upload a job description: Claude Sonnet 4.5 parses it into the actual skills and stories the hiring manager will dig into
- Generate a personalized story bank from your resume: not generic STAR templates, your real experiences shaped to the role
- Gap analysis: Gemini 2.5 Flash finds the holes between what the JD wants and what your stories cover, then generates the questions you're most likely to get asked
- Voice reps: record your answer, Whisper transcribes, Claude evaluates the actual audio for specificity, rambles, and confidence
- Stripe payments: $4.99 per use or $19/month for twenty packages, no free-trial roulette, no enterprise sales motion
Live with paying users on both tiers. A candidate goes from job description to coached voice rep in under ten minutes, without a human coach in the room.
Next.js 16 · App Router · Vercel
Supabase: Postgres, RLS, email/password
Claude Sonnet 4.5: JD parsing, answer generation, audio evaluation
Gemini 2.5 Flash: gap analysis, question generation
OpenAI Whisper: transcription only
Stripe live: per-use and subscription
Tailwind CSS 4 · dark mode first
The story of speakeasyprep is the story every AI-curious founder should know: no-code lets you earn the right to build, but it won't carry paying users. The judgment call is knowing which week to switch, and what to keep from the prototype when you do. That judgment is what you're hiring for when you hire Barael.