AI Application Helper

This project orchestrates multi-agent workflows for automating application drafting and submission using LLMs and browser automation. It features:

Browser Automation

This project uses browser-use for browser automation. The submission agent leverages browser-use to:

• Open and interact with web forms
• Fill out and submit applications automatically
• Extract content from success pages after submission

Make sure to follow the browser-use setup instructions for your platform. The browser-use package is a core dependency for the submission agent's automation capabilities.

• Multi-agent workflow for data ingestion, research, writing, validation, and submission
• MCP server exposing the submission agent as a tool
• LangChain/LangGraph integration for agent-tool orchestration
• Streamlit UI for user interaction

Features

• Summarize user profile and target entity
• Draft and validate applications (SOP, job, etc.)
• Submit applications via browser automation
• Modular agents for each workflow step
• MCP server for tool interoperability

Structure

main.py                  # Entry point
ui.py                    # Streamlit UI
requirements.txt         # Dependencies
agents/                  # Agent implementations
workflows/               # Workflow orchestration
submission_mcp_server.py # MCP server for submission agent

Usage

1. Install dependencies

pip install -r requirements.txt

2. Set up environment variables

Create a .env file with your API keys (e.g., for Gemini):

GEMINI_API_KEY=your_key_here

3. Run the MCP server (for submission agent)

python submission_mcp_server.py

4. Run the Streamlit UI

streamlit run ui.py

5. (Optional) Use the SubmissionAgent directly

You can invoke the submission agent as a tool from other agents or scripts using the MCP protocol.

Notes

Example Workflow

1. User provides profile sources and a target entity (e.g., university or company).
2. The workflow loads and summarizes the profile data.
3. Entity research agent gathers and summarizes information about the target.
4. Writing agent drafts an application (e.g., SOP) using the summaries.
5. Validation agent reviews the draft for quality and relevance.
6. Submission agent (via MCP server) fills and submits the application form using browser automation.

Example Python usage:

from workflows.application_workflow import ApplicationWorkflow

workflow = ApplicationWorkflow(gemini_api_key="your_gemini_key")
profile_sources = ["data/my_cv.pdf"]
user_prompt = "Apply to the MS program at Example University."
submission_url = "https://exampleuniversity.edu/apply"

result = workflow.run(profile_sources, user_prompt, submission_url)
print(result["application"])
print(result["submission_result"])

---

• Make sure only one MCP server instance runs on a given port.
• All logs and print statements in the MCP server should go to stderr, not stdout.
• The project supports both direct tool invocation and agent-based orchestration.