🎯 AI Job Application Agent

An AI-powered assistant that manages a full job search end-to-end: track applications, generate tailored cover letters, analyse job–CV fit, ingest and classify recruiter emails, and prepare for interviews — all from one local Streamlit app.

🧪 New — CV & Role Finder: upload a CV → get recommended roles to target → screen a job for fit. Works offline (local matching) or bring-your-own API key. streamlit run ui/cv_finder.py — this is the first slice of a planned multi-user version.

Built as a personal project to explore multi-provider LLM orchestration, the Model Context Protocol (MCP), and a clean data layer around a real-world workflow.

Privacy note: this repo ships no personal data. Your applications, cover letters, CV and API keys live in git-ignored files. A one-command demo seeds realistic fake data so you can try it immediately.

✨ Features

• Multi-provider AI — Groq, Google Gemini, OpenAI, with an offline keyword fallback and an auto mode that picks the best available provider.

• Cover-letter generation tailored to a job description + your CV summary.

• Job–CV fit analysis — score, matched skills, gaps, recommendation.

• Application tracker — SQLite-backed, with a full change-history audit trail.

• Email intelligence — scan a Gmail inbox, classify messages (rejection / interview / offer / scheduling) with an LLM, and auto-update statuses.

• Interview prep — practice questions, company research, feedback tracking.

• Calendar export — interviews to .ics.

• MCP server — exposes the agent's tools over the Model Context Protocol.

• Streamlit UI — Dashboard, Applications (table / cards / kanban), Email, CV & Insights, Settings.

Related MCP server: RSpace MCP Server

🏗️ Architecture

┌────────────┐     ┌──────────────────┐     ┌───────────────────────────────┐
│ Streamlit  │────▶│ ApplicationAgent │────▶│ LLM providers                 │
│ UI (ui/)   │     │ (agent/)         │     │ groq · gemini · openai · local│
└─────┬──────┘     └──────┬───────────┘     └───────────────────────────────┘
      │                   │
      ▼                   ▼
┌────────────┐     ┌──────────────────┐     ┌──────────────┐
│ tools/     │     │ db/ (SQLAlchemy) │────▶│ SQLite       │
│ email·jobs │     │ models·session   │     │ applications │
└────────────┘     └──────────────────┘     └──────────────┘
      ▲
      │
┌────────────┐
│ mcp_server/│  Model Context Protocol tools
└────────────┘

🚀 Quick start

# 1. Create the environment (Python 3.13)
python3.13 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

# 2. Configure secrets
cp .env.example .env         # then add your API keys (all optional; "local" needs none)

# 3a. Try it with demo data (recommended first run)
python scripts/seed_demo_data.py
APP_DB_PATH=applications_demo.db streamlit run ui/app.py

# 3b. …or run against your own data
cp data/cv_summary.example.txt data/cv_summary.txt   # then edit with your CV
./run_ui.sh

App opens at http://localhost:8501.

⚙️ Configuration

All configuration is via environment variables (see .env.example):

📁 Project layout

agent/        Multi-provider AI agent (cover letters, fit analysis)
ui/           Streamlit app — page functions + components
tools/        Email tracking/analysis, job scraping & search
db/           SQLAlchemy models, session, migrations
mcp_server/   MCP server exposing agent tools
scripts/      Utilities (e.g. seed_demo_data.py)
cli/          Command-line interface
data/         Local data (git-ignored; *.example.* files are shipped)

🛠️ Tech stack

Python 3.13 · Streamlit · SQLAlchemy + SQLite · OpenAI / Groq / Google Gemini SDKs · Model Context Protocol · pandas · scikit-learn

🗺️ Roadmap

• Unify provider logic behind a single LLMProvider interface

• Structured logging (replace prints) + pytest test suite + CI

• Migrate google-generativeai → google-genai

• Dockerfile + Compose for one-command run and deployment

• Semantic (embedding-based) job–CV fit scoring

📄 License

MIT © 2026 Philipp Goetting