Skip to content

Ceevee

AI resume optimizer

Category: Product & client work

  • Built with:
  • TypeScript
  • Next.js 16
  • React 19
  • Tailwind v4
  • PostgreSQL
  • Prisma
  • Redis
  • LangChain.js
  • Claude API
  • Zod
  • Puppeteer
  • html-to-docx
  • MinIO
  • Docker
  • Paddle
  • JWT
The Ceevee optimizer showing an ATS score of 76 out of 100, matched and missing keyword chips, and a ranked recommendation list led by the skills the evidence guard removed.

What it is

Ceevee rewrites a structured master resume against any job posting — returning an ATS-scored resume, cover letter, interview prep and salary estimate. Every skill the model claims is checked against evidence in that record, and anything unsupported is deleted and reported.

Truthfulness is enforced in code, not in the prompt.

Value

  • A higher score never buys a false claim. Keyword optimizers reward lying. The guard runs on every generation: the score goes down, and the reason is on screen.
  • The gaps become the product. Removals feed missing keywords, ranked recommendations, and interview questions that rehearse the weakness.
  • Cost is a design constraint. Cache breakpoints, a stripped candidate payload and an optional-schema cover letter keep spend per run predictable.

Technology highlights

  • Evidence guard — deterministic, and conservative. Tokenizes the whole master resume recursively. Node.js = NodeJS = node-js, but word boundaries hold — Java is never satisfied by JavaScript.
  • Claude pipeline — engineered for cost and failure. Eight chains, schema bound as a tool rather than parsed. A cache breakpoint splits the user message; retries split 3 transport / 1 validation.

More projects