TailorLoop

AI-powered job application copilot. Paste a job description, get a tailored résumé and cover letter PDF — every claim verified against your real experience, nothing fabricated.

How it works

A LangGraph pipeline runs four agents against your profile and the job description:

1. jd_analyzer — extracts required skills, ATS keywords, and responsibilities
2. resume_tailor + cover_letter_writer — select and rephrase real entries from your content bank (run in parallel)
3. verifier — checks every claim traces back to your profile; rejects anything invented
4. compile — renders approved content to LaTeX and compiles to PDF via Tectonic

If the verifier rejects claims, the writers retry with the rejection reasons. After 2 retries, rejected content is dropped rather than kept.

---

Prerequisites

# uv — Python package manager
curl -LsSf https://astral.sh/uv/install.sh | sh

# Tectonic — LaTeX → PDF compiler
brew install tectonic

# Node.js 18+ — for the frontend
# https://nodejs.org

---

Setup

1. Clone and install Python dependencies

git clone https://github.com/Rutvik2598/TailorLoop.git
cd TailorLoop
uv sync

2. API key

Create a .env file in the project root:

echo 'GOOGLE_API_KEY=your-key-here' > .env

Get a key at aistudio.google.com. The .env file is gitignored.

3. Install frontend dependencies

cd web && npm install && cd ..

---

Running the app

Step 1 — Initialise the database

Run once to create tailorloop.db and seed it with the profile data from profile/:

uv run --env-file .env tailorloop seed --profile profile/ --db tailorloop.db

To wipe and re-seed at any point, delete tailorloop.db and run the command again.

Step 2 — Start the API server

uv run --env-file .env tailorloop serve --db tailorloop.db

Server runs at http://127.0.0.1:8000. Interactive API docs at http://127.0.0.1:8000/docs.

Step 3 — Start the frontend

cd web && npm run dev

Open http://localhost:3000. The Generate page is the home screen — paste a job description and hit Generate.

---

CLI usage (without the web UI)

Run the pipeline directly against a JD file:

uv run --env-file .env tailorloop run \
  --jd examples/jd_backend_engineer.txt \
  --db tailorloop.db \
  --out out/

Outputs:

out/
├── resume.pdf
├── cover_letter.pdf
└── run.json        ← full audit trail (every agent input/output)

Flags

Flag	Default	Description
--db	tailorloop.db	Path to SQLite database
--out	out/	Output directory for PDFs and audit log
--max-retries N	2	Verifier retry cap
--no-cache	off	Disable on-disk LLM response cache
-v	off	Verbose structured logging

---

Project structure

TailorLoop/
├── src/tailorloop/
│   ├── agents/          # jd_analyzer, resume_tailor, cover_letter_writer, verifier
│   ├── api/             # FastAPI app + generate job runner
│   ├── db/              # SQLite schema, CRUD, seeding
│   ├── llm/             # Model abstraction (get_model by role) + response cache
│   ├── prompts/         # One file per agent prompt
│   ├── graph.py         # LangGraph state graph wiring
│   ├── render.py        # ResumeSelection/CoverLetterDraft → LaTeX
│   ├── compile.py       # LaTeX → PDF via Tectonic
│   ├── models.py        # Pydantic models (Profile, RunState, etc.)
│   ├── loaders.py       # Load profile from DB or from JSON files
│   └── cli.py           # CLI commands: run, seed, serve
├── templates/           # resume.tex.j2, cover_letter.tex.j2
├── profile/             # Seed data JSON files
├── examples/            # Sample job descriptions
├── tests/               # Fabrication regression tests
├── web/                 # Next.js frontend
│   ├── app/             # Pages: generate, about, experience, projects, education, skills
│   ├── components/      # Sidebar, shadcn/ui components
│   └── lib/             # Typed API client
├── .env                 # GOOGLE_API_KEY (gitignored)
├── tailorloop.db        # SQLite database (gitignored)
└── pyproject.toml

---

Stack

Backend: Python 3.11 · LangGraph · Pydantic v2 · FastAPI · SQLite · Google Gemini (google-genai) · Tectonic · Jinja2 · uv

Frontend: Next.js · TypeScript · Tailwind CSS · shadcn/ui