Resume Cartel

Resume Cartel is a Google ADK (Agent Development Kit) app that builds role-targeted resumes with a coordinated team of agents—coach, writer, critic, and YAML export—so you ship evidence-backed drafts instead of generic filler.

Features

• Multi-Agent Architecture: Specialized agents for coaching, writing, critiquing, and formatting.
• Interactive Coaching: A user-facing agent that gathers requirements and guides you through the process.
• Quality Loop: Built-in critic agent ensures high-quality evidence, role fit, and consistency.
• Markdown Artifacts: Maintains versioned working documents (e.g., candidate-profile.md, resume-draft.md).
• YAML Export: Automatically converts the finalized markdown resume into a structured resume.yaml format.
• Integrated Research: Agents have access to Google Search for real-time research.

Example session

The example/ directory includes conversation screenshots and a frozen snapshot of ADK session artifacts so you can see how a real run progresses without starting the app.

Conversation flow

1. First-time interaction — The coach asks focused questions; you supply the role you are targeting, background, constraints, and any evidence you want emphasized. That input is distilled into working documents (for example requirements.md, raw-notes.md, and candidate-profile.md).

2. Critic–writer loop and draft — After enough context is captured, the writer and critic agents iterate: the writer updates resume-draft.md, the critic records feedback in critique.md, and the coach drives further revisions until the draft stabilizes (subject to RESUME_MAX_REVISION_ITERATIONS).

Example session artifacts

Under example/adk_session_artifacts/ you will find the same artifact names as in Working Documents, each stored under versions/<n>/ the way ADK persists them during a session. Browse those files to follow how requirements become a profile, then a critiqued draft, and optionally exports such as final-resume.md and resume.yaml.

Architecture

graph TD
    User([User]) <-->|Interacts| Coach[ResumeCoachAgent]
    Coach -->|Delegates| Writer[ResumeWriterAgent]
    Coach -->|Delegates| Critic[ResumeCriticAgent]
    Coach -->|Delegates| Yaml[ResumeYamlAgent]
    
    Writer -.->|Drafts & Revises| MD[(Markdown Artifacts)]
    Critic -.->|Reviews| MD
    Yaml -.->|Converts| YAML[(resume.yaml)]
    Coach -.->|Manages State| MD

Loading

Prerequisites

• Python 3.13+
• uv (Python package manager)

Environment Setup

Create a .env file in the root directory with the following variables:

# Required: Google Gemini API Key
GEMINI_API_KEY=your_api_key_here

# Optional: Model overrides (defaults to gemini-3.1-pro-preview)
RESUME_COACH_MODEL=gemini-3.1-pro-preview
RESUME_WRITER_MODEL=gemini-3.1-pro-preview
RESUME_CRITIC_MODEL=gemini-3.1-pro-preview

# Optional: Max revisions before forcing finalization
RESUME_MAX_REVISION_ITERATIONS=4

How to Run

1. Install dependencies and start the ADK web interface:

uv run adk web

2. The ADK app entrypoint is resume_builder.agent:root_agent. Open the provided local URL in your browser to start a Resume Cartel session with the coach agent.

Working Documents

The system uses the following ADK artifacts as versioned working documents during the session:

• requirements.md
• raw-notes.md
• coach-notes.md
• candidate-profile.md
• resume-draft.md
• critique.md
• final-resume.md
• resume.yaml (Generated upon request)