khetansarvesh/job-cv-tailor

ATS-Optimized CV Tailor (LaTeX)

Generate a tailored CV for a specific job description. Reads your LaTeX resume template, edits it to match the JD (rewrite summary, reorder bullets, select projects, inject keywords), saves a tailored copy, and compiles to PDF locally.

When to Activate

• User asks to generate a CV or resume for a specific role
• User wants to tailor their resume to a JD
• User says "make a PDF", "generate CV", or "tailor my resume"
• After running job-eval, user wants a customized CV for a high-scoring role

Prerequisites

• profile/resume.tex — LaTeX resume template (source of truth — never overwrite this)
• Notion (fetch via python3 scripts/notion/page_reader.py resume) — structured experience, education, skills data
• Notion (fetch via python3 scripts/notion/page_reader.py projects) — project summaries
• pdflatex installed locally (TeX Live or MacTeX)
  • macOS: brew install --cask mactex or brew install --cask basictex
  • Lato font: tlmgr install lato
• A job description (text, URL, or from a previous evaluation)

Non-Negotiables

1. Never invent skills or experience — only reformulate what actually exists in the profile
2. Every bullet must trace to real experience — keyword injection means rephrasing, not fabricating
3. Single page — if the tailored version exceeds 1 page, trim content (fewer projects, shorter bullets)
4. Preserve LaTeX structure — don't break macros, formatting commands, or section ordering
5. Never auto-submit the CV — user reviews the final PDF

Workflow

Step 1: Read JD and Profile

1. Get the job description (user pastes text, or use WebFetch on a URL)
2. Read profile/resume.tex — the LaTeX template you'll modify
3. Read Notion (fetch via python3 scripts/notion/page_reader.py resume) and Notion (fetch via python3 scripts/notion/page_reader.py projects) for full context
4. If an evaluation report exists (from job-eval), read it for archetype and keyword analysis

Step 2: Extract Keywords (15-20)

From the JD, extract:

• Hard skills: specific technologies, frameworks, tools, languages
• Soft skills: leadership patterns, collaboration styles
• Domain terms: industry-specific vocabulary
• Action verbs: what the role does (build, deploy, research, optimize, lead)

Rank by frequency and prominence in the JD. The top 5 go in the summary.

Step 3: Detect Archetype

Classify the role to frame the summary appropriately:

| Archetype | JD Signals | Summary Emphasis | | ----------------- | ---------------------------------------------------- | -------------------------------------------------------------- | | AI/ML Research | "research", "SOTA", "publications", "benchmarks" | SOTA results, publications, experiment methodology | | LLM/NLP | "RAG", "agents", "LLM", "prompting", "orchestration" | RAG pipelines, agent orchestration, LLM optimization | | ML Platform/MLOps | "pipelines", "serving", "monitoring", "CI/CD" | Production systems, observability, evals, deployment | | Computer Vision | "vision", "detection", "diffusion", "VLM", "video" | Detection, generation, VLMs, diffusion models | | Data Science | "analytics", "experimentation", "modeling", "A/B" | Business impact, experimentation, stakeholder communication | | Applied AI | "end-to-end", "prototype", "production", "deploy" | End-to-end delivery, cross-functional, prototype-to-production |

Step 4: Plan Tailored Content

Before editing any LaTeX, decide what changes to make:

Professional Summary (2-4 sentences):

• Inject top 5 JD keywords naturally
• Lead with strongest match point for the detected archetype
• Include one quantified achievement
• Bridge from your background to this specific role

Technical Skills reorder:

• Within each category (Languages, Frameworks, Cloud, Dev Tools, IT Constructs), move JD-matching items to the front

Work Experience bullet reorder:

• For each job entry, score bullets by JD relevance (0-3)
• Put most relevant bullets first
• Optionally rephrase 1-2 top bullets to incorporate JD vocabulary

Projects selection:

• From the 7 project bullets, select top 4-5 most relevant
• Reorder by relevance
• Optionally rephrase to emphasize JD-matching aspects

Achievements filter (if needed for space):

• Keep the most relevant 3-4 of 5

Step 5: Keyword Injection Rules (Ethical)

ALLOWED:

• Reformulate existing experience using JD vocabulary
  • "LLM workflows with retrieval" → "RAG pipeline design and LLM orchestration"
• Reframe true skills with JD terminology
  • "observability, evals, error handling" → "MLOps: observability, evals, cost monitoring"
• Elevate stakeholder context
  • "collaborated with team" → "stakeholder management across engineering and business"

FORBIDDEN:

• Adding skills the user doesn't have
• Inventing metrics or experience
• Claiming expertise in unused technologies

Validation: Every keyword must trace back to Notion (fetch via python3 scripts/notion/page_reader.py resume) or Notion (fetch via python3 scripts/notion/page_reader.py projects).

Step 6: Edit resume.tex

Read profile/resume.tex and make modifications using the Edit tool:

6a. Add Professional Summary

Insert a new section after the header and before Education:

%-----------PROFESSIONAL SUMMARY-----------
\section{Professional Summary}
\begin{itemize}[leftmargin=0.2in, label={}, nosep]
  \item \small{
    [2-4 sentences with top 5 JD keywords, archetype framing, one quantified achievement]
  }
\end{itemize}

If a Professional Summary section already exists (from a previous tailoring), replace its content.

6b. Reorder Technical Skills

Find the Technical Skills section. Within each \textbf{Category:} line, reorder the comma-separated items to frontload JD-matching keywords. Example:

Before: \textbf{Frameworks:} PyTorch, Hugging Face, Langchain, LangGraph, LlamaIndex, PySpark... After (if JD emphasizes RAG/agents): \textbf{Frameworks:} Langchain, LangGraph, LlamaIndex, PyTorch, Hugging Face, PySpark...

6c. Reorder Work Experience Bullets

For each job entry, reorder the \resumeItem{...} lines. The most JD-relevant bullet goes first. Optionally rephrase 1-2 bullets to weave in JD vocabulary naturally.

Example — if JD emphasizes "multi-agent systems": Move the SERA bullet (multi-agent, 40+ tools) above the Harbor ATIF bullet.

6d. Select & Reorder Projects

From the 7 \resumeItem{...} lines in the Projects section, keep only the 4-5 most relevant. Remove the rest. Reorder by JD relevance.

6e. Preserve Everything Else

Do NOT modify:

• Header/contact info
• Education section
• LaTeX preamble/macros
• Section ordering (Education → Skills → Work Exp → Publications → Projects → Achievements)
• Hyperlinks
• Formatting commands

Step 7: Save Modified .tex

Save the tailored version as a new file (never overwrite the original):

profile/resume_tailored_{company-slug}.tex

Where {company-slug} is the company name in lowercase with hyphens (e.g., scale-ai, anthropic, cohere).

Step 8: Compile to PDF

Run via Bash:

cd /path/to/ai_skills_repo/profile && pdflatex -interaction=nonstopmode resume_tailored_{company-slug}.tex

Run twice (LaTeX sometimes needs two passes for proper rendering):

cd /path/to/ai_skills_repo/profile && pdflatex -interaction=nonstopmode resume_tailored_{company-slug}.tex && pdflatex -interaction=nonstopmode resume_tailored_{company-slug}.tex

Then move the PDF to the output directory:

mv profile/resume_tailored_{company-slug}.pdf applications_database/resumes/cv-{company-slug}-{YYYY-MM-DD}.pdf

Clean up LaTeX auxiliary files:

rm -f profile/resume_tailored_{company-slug}.{aux,log,out}

Step 9: Verify & Present

1. Check compilation succeeded — exit code 0 and PDF file exists
2. Check page count — warn if PDF is >1 page (content may need trimming)
3. Present summary to user:

CV tailored for: {Company} — {Role}
Archetype: {detected}
Keywords injected: {list of top keywords}
Changes made:
  - Summary: added/updated with {keywords}
  - Skills: reordered {categories}
  - Experience: reordered bullets in {job entries}
  - Projects: selected {N} of 7, removed {removed ones}
Output: applications_database/resumes/cv-{company-slug}-{date}.pdf

Cover Letter Generation

When a cover letter is required (form marks it mandatory, or user asks):

1. Structure (1 page max, 4 paragraphs):
   • Paragraph 1: Why this company specifically (concrete fact, not flattery)
   • Paragraph 2: Your strongest match point with a quantified result
   • Paragraph 3: Bridge — how your background connects to what they need (archetype-aware)
   • Paragraph 4: Forward-looking close
2. JD keywords woven naturally throughout
3. Save as plain text or compile a separate LaTeX cover letter

Error Handling

| Issue | Solution | | ----------------------- | ---------------------------------------------------------------------------------- | | pdflatex not installed | Tell user: "Install TeX Live: brew install --cask mactex" | | Compilation error | Read the .log file for errors, fix LaTeX syntax issues | | PDF > 1 page | Trim: reduce projects from 5→3, shorten bullets, remove least relevant achievement | | Font not found (Lato) | Tell user: "Run sudo tlmgr install lato" | | Original .tex corrupted | Never happens — we never overwrite profile/resume.tex |

File Structure

profile/
├── resume.tex                              # Original template (NEVER overwrite)
├── resume_tailored_scale-ai.tex            # Tailored copy for Scale AI
├── resume_tailored_anthropic.tex           # Tailored copy for Anthropic
└── ...

applications_database/
└── resumes/
    ├── cv-scale-ai-2026-04-16.pdf          # Compiled tailored PDFs
    ├── cv-anthropic-2026-04-16.pdf
    └── ...

Related Skills

• job-eval — Evaluate the offer first (provides archetype + keywords for smarter tailoring)
• job-apply — Fill the application form after generating the CV
• job-interview-prep — Prepare for interviews at this company
• job-tracker — Track the application status