Skilltune: Agentic Resume Tailoring via Chrome Extension

Skilltune is a Chrome extension that reads job postings in the browser, analyzes them with AI agents, and generates a tailored .typ (Typst) resume compiled to PDF — all without leaving the page.

How It Works

1. Onboarding — Enter your name, target roles, industry, and path to your base resume file (.typ).
2. Scan — Browse a job posting, open the extension popup, click “Scan this page” to capture the page text.
3. Analyze — Click “Analyze posting” to have an AI agent extract job title, company, skills, requirements, and responsibilities.
4. Write — Click “Write tailored resume” to generate a customized .typ file using your resume as a style reference, compiled to PDF via Typst CLI.

Architecture

┌─────────────────────┐  chrome.runtime      ┌───────────────────────┐
│   Chrome Extension   │ sendMessage(API_CALL)  │   Python Server        │
│                     │ ──────────────────────→ │   (localhost:3721)     │
│  Popup (React)      │ ←────────────────────── │                        │
│  Service Worker     │     JSON response       │  LangChain Agents:     │
│  Content Script     │                         │  - posting-analysis   │
│                     │                         │  - resume-generation  │
└─────────────────────┘                         └───────────────────────┘

Chrome Extension Components

• Popup (React) — User interface for scanning, analyzing, and generating
• Service Worker — Background script handling API communication
• Content Script — Captures page text from job posting tabs

Python Backend Agents

• Posting Analysis Agent — Extracts structured data (title, company, skills, requirements, responsibilities) from raw page text using LangChain + DeepSeek.
• Resume Generation Agent — Takes the extracted job data and the user’s base .typ resume, then generates a tailored version that emphasizes relevant skills and experiences.

Typst as Resume Format

Typst was chosen over LaTeX or Word for several reasons:

• Fast compilation — Typst CLI compiles to PDF in milliseconds
• Clean syntax — More readable than LaTeX, easier for LLMs to generate correctly
• Programmatic control — The agent can modify sections, reorder bullet points, and adjust emphasis while maintaining consistent formatting
• Local rendering — No cloud rendering needed; typst compile runs locally

Key Design Decisions

1. Local Python server over cloud API — Keeps resume data on-device, avoids sending personal documents to third parties.
2. Two-agent pipeline — Separation of concerns: one agent analyzes the job, another generates the resume. This allows independent improvement and testing.
3. Screenshot-based page capture — More reliable than DOM scraping for complex job board layouts.
4. Style reference preservation — The generation agent uses the existing .typ file as a template, ensuring consistent formatting across all tailored resumes.

Tech Stack

• Extension: React, Chrome Extension APIs (runtime, tabs, scripting)
• Backend: Python, LangChain, DeepSeek API
• Document Format: Typst (.typ → PDF via typst compile)