ResumeForge AI

An AI-powered resume tailoring system that analyzes job descriptions and generates a customized, ATS-optimized LaTeX resume — ready to compile and submit.

How it works

Three specialized AI agents run sequentially:

Agent	Model	Role
Job Analyst	Gemini 2.5 Flash (free tier)	Extracts skills, requirements, tone, and cultural signals from the job description
Resume Tailor	Qwen3:14b (local via Ollama)	Selects the most relevant projects from your pool, rewrites bullets with job-specific keywords, and enforces a one-page budget
Resume Validator	Gemini 2.5 Flash (free tier)	Scores ATS compatibility, keyword integration rate, and human readability before generating output

Resume tailoring runs fully locally (Ollama) — your resume data never leaves your machine for the creative step. Gemini free tier handles the structured extraction tasks.

Output

• LaTeX file saved to resumestex/{Company}_{Role}.tex — compile with pdflatex or Overleaf
• Validation report printed to terminal with ATS score, keyword integration rate, and suggestions
• Application log added to your Notion database with the full job description saved for interview prep

Setup

1. Prerequisites

• Python >= 3.10
• uv — pip install uv
• Ollama with qwen3:14b pulled: ollama pull qwen3:14b

2. Install

git clone https://github.com/rishikesh-20/resumeforge.git
cd resumeforge
uv sync

3. Environment variables

Copy the example and fill in your keys:

cp .env.example .env

Then edit .env:

GOOGLE_API_KEY=your_google_ai_studio_key
NOTION_API_KEY=your_notion_integration_key
NOTION_DATABASE_ID=your_notion_database_id

Getting these keys:

• GOOGLE_API_KEY — Google AI Studio → Get API key (free)
• NOTION_API_KEY — notion.so/my-integrations → New integration → copy secret
• NOTION_DATABASE_ID — The ID from your Notion database URL: notion.so/{workspace}/{DATABASE_ID}?v=...

4. Notion database setup

Step 1 — Create an integration

Go to notion.so/my-integrations → New integration → give it a name → Submit. Copy the Internal Integration Secret — this is your NOTION_API_KEY.

Step 2 — Create a database

Create a new full-page database in Notion with exactly these properties:

Property	Type
Name	Title
Company	Text
Role	Text
Location	Text
Date Applied	Date
Status	Select — add options: Applied, Interview, Offer, Rejected
Job ID	Text

Step 3 — Get the database ID

Open the database in Notion. The URL looks like:

https://www.notion.so/{workspace}/{DATABASE_ID}?v=...

Copy the DATABASE_ID part (32-character string before ?v=) — this is your NOTION_DATABASE_ID.

Step 4 — Connect the integration to the database

Open the database → click ... (top right) → Connections → search for your integration → click Connect. Without this step, the API key alone won't work.

5. Add your resume data

Copy the example and fill in your details:

cp data/resume.json.example data/resume.json

Edit data/resume.json with your actual header, work experience, education, and skills.

6. Add your projects

Projects live in data/projects/ as individual JSON files — one file per project. Copy the example to get started:

cp data/projects/project.json.example data/projects/my-project.json

Add as many projects as you have. For each run, the AI scores all projects against the job description and selects the top 3 most relevant ones. The description field should be a detailed paragraph — the more context you give, the better the AI can tailor the bullet points.

Example project file:

{
  "name": "My Project",
  "tech_stack": "Python, FastAPI, PostgreSQL",
  "date": "December 2024",
  "description": "Detailed paragraph about what you built, why, how, and any results or metrics..."
}

Usage

Web app (Streamlit)

uv run streamlit run app.py

Fill in the company, job title, location, optional job ID, and paste the job description. The app shows pipeline progress, the validation report, and a preview of the generated LaTeX, with a button to download the .tex file. The file is also saved to resumestex/ and the application is logged to Notion, same as the CLI.

CLI

uv run resumeforge

You will be prompted for:

• Company name, job title, location, job ID (optional)
• Job description (paste and type END on a new line to finish)

The three agents run sequentially (~3–6 minutes total, mostly the local Ollama step), then the .tex file is saved to resumestex/ and the application is logged to Notion with the full job description.

Compiling to PDF

Once you have a .tex file:

pdflatex resumestex/Company_Role.tex

Or paste the contents into Overleaf for online compilation.

Customizing the LaTeX template

The template is at templates/resume_latex_template.tex. It uses Jinja2 with (( )) for variables and (% %) for blocks (to avoid conflicts with LaTeX syntax). Edit the static sections (certifications, formatting, margins) directly. Preserve the (( )) and (% %) placeholders.

Project structure

resumeforge/
├── data/
│   ├── resume.json               # Your master resume (gitignored — copy from resume.json.example)
│   ├── resume.json.example       # Template to copy and fill in
│   └── projects/                 # One JSON file per project (gitignored)
│       └── project.json.example  # Template to copy for each project
├── resumestex/                   # Generated .tex files (gitignored)
├── templates/
│   └── resume_latex_template.tex # Jinja2 LaTeX template
└── src/resumeforge/
    ├── config/
    │   ├── agents.yaml           # Agent roles and backstories
    │   └── tasks.yaml            # Task instructions
    ├── utils/
    │   ├── resume_latex_generator.py
    │   ├── notion_tracker.py
    │   ├── read_projects.py      # Loads all projects from data/projects/
    │   └── ...
    ├── agents.py                 # Agent and task definitions
    ├── models.py                 # Pydantic data models
    └── main.py                   # Entry point

Built with

• Qwen3:14b via Ollama — local LLM for resume tailoring
• Gemini 2.5 Flash — free-tier LLM for analysis and validation
• Jinja2 — LaTeX template rendering
• notion-client — application tracking