How I built a daily job application pipeline with AI

The problem

Job hunting for AI automation and performance marketing roles means trawling 8+ job boards, reading each description, rewriting your resume to match, and repeating that for every application. Most people send the same generic resume everywhere or burn hours tailoring by hand. Neither works well.

The architecture

JobTailor is a factory line with five stages. Each stage does one thing and hands the result to the next.

Stage 1: Find jobs

The pipeline searches 8+ sources: Adzuna, Wellfound, Glassdoor, foundit, SimplyHired, DailyRemote, LinkedIn, and company career pages. Everything is filtered to remote-only. Each source is toggleable in the config. The results are deduplicated by URL.

Stage 2: Read job descriptions

For each job URL, the pipeline pulls the full description and extracts the skills and requirements the employer wants. It uses Scrapling, a Python web scraping framework with anti-bot bypass, to handle sites that block automated access.

Stage 3: Rewrite the resume

This is the core. An LLM reads the job description and adjusts a LaTeX resume (Awesome-CV template) to highlight matching skills. It reorders skills sections, rephrases experience bullets to surface relevant keywords, and adjusts the position line.

The guardrail: the AI never invents skills or experience. It only rephrases and reorders what is already on the resume. Nothing ships that you could not defend in an interview. Each tailored resume lives on its own git branch. The original is never touched.

Stage 4: Compile PDF

Each rewritten LaTeX file is compiled into a PDF using Tectonic, a self-contained LaTeX compiler. No full TeX installation needed. The output is a clean, formatted PDF ready to upload.

Stage 5: Open applications

The pipeline opens each job's application page with the tailored PDF pre-attached. You review each application and click Submit. Human-in-the-loop by default. No auto-submission that could get your account banned.

The tech stack

• Python package with five pipeline stages
• Scrapling for anti-bot job scraping
• LLM API for resume tailoring (OpenAI-compatible)
• Tectonic for LaTeX to PDF compilation
• Next.js + FastAPI dashboard for monitoring runs
• ego-browser for application auto-fill
• macOS LaunchAgent for 7am daily scheduling

Results

End-to-end verified: 10 leads in, 2 job descriptions matched, 2 tailored PDFs out, 2 application pages opened. The pipeline runs daily and hands me a stack of ready-to-submit applications every morning. What used to take 2 hours of manual work now takes 15 minutes of review.

The hardest part

Not the AI. Not the scraping. The resume tailoring. Getting an LLM to rewrite a resume without inventing skills is harder than it sounds. The prompt has to be explicit: rephrase, reorder, highlight. Do not add. Do not embellish. The model will try to make the resume sound better by adding claims the user never made. The guardrail is in the prompt and in the validation step.

The project is open source at github.com/mueedbhat-create/jobtailor. Anyone can install it, point it at their own resume and keywords, and run their own automated job search. For more on AI automation in marketing, see my AI marketing automation guide and my n8n beginner tutorial.