AI application · Python

Callback

Writes a tailored resume and cover letter for a job in the user's own voice, with Claude doing narrow, testable tasks inside a Django, Celery and Next.js stack.

The problem

Tailoring a resume and cover letter to each job is slow. Generic AI rewrites are fast, but they invent experience and do not sound like the person applying.

What I built

Callback takes a resume, a sample cover letter and optional supporting documents, plus a pasted job description, and writes a resume and cover letter for that job in the user's own voice. The user edits both in the app with a live preview and downloads PDFs.

The frontend is Next.js. The backend is Django and Django REST Framework with Celery and Redis for the generation jobs, Postgres for data and full-text search, and WeasyPrint for the PDFs.

How AI was used

Generation runs in three stages with one Claude model each: a context scout that uses tools to find the relevant parts of the user's documents, with a cap on its steps; a writer; and a verifier that flags claims the source documents do not support. Retrieval is Postgres full-text search, with no embeddings.

The repository carries a written specification for AI coding agents, so that agents working on the code stay within the architecture.

What kept it safe

  • Unsupported claims are shown to the user in a flagged-claims panel instead of being silently kept.
  • Supporting documents are strictly scoped to their owner.
  • Authentication uses httpOnly cookies. Uploads and content are validated, and layout values are checked before they reach CSS.
  • Parsing, generation and sign-in are throttled, guarded against duplicate in-flight requests and capped per day, because each generation costs money.

Results

The first version is merged: paste a job description, get both documents, edit them and export PDFs. It has been verified end to end against the real Anthropic API and in a headless browser, using a fictional resume. It is not deployed yet.

Lessons and technologies

  • Keep the language model on narrow tasks and put testable code around it.
  • Next.js, Django, Django REST Framework, Celery, Redis, PostgreSQL, Anthropic API, WeasyPrint, AWS S3, Docker.