AI Article Generator

Generate a structured, illustrated article from a single keyword. Pick a topic, pick the sections you want, and the app writes the introduction, each section, the conclusion, and a cover image using the OpenAI API.

Stack: Django 5.2 + Django REST Framework · React 19 + Vite 7 + Tailwind CSS 3

The application is stateless — it defines no models and stores no user content. Generated articles are returned in the HTTP response and never persisted.

---

Quick start

Requires Python 3.12+ and Node 22 (see frontend/.nvmrc). No database server is needed: the default configuration uses SQLite, which Django creates on first run.

Backend

cd backend
python -m venv .venv
source .venv/bin/activate        # Windows: .venv\Scripts\activate
pip install -r requirements.txt

cp .env.example .env             # then set OPENAI_API_KEY and DEBUG=True
python manage.py migrate
python manage.py runserver

The API is now on http://localhost:8000/api/.

Frontend

cd frontend
npm ci
cp .env.example .env             # defaults already point at localhost:8000
npm run dev

The UI is now on http://localhost:5173.

Docker Compose

cp backend/.env.example backend/.env
# set OPENAI_API_KEY and DJANGO_SECRET_KEY in that file
docker compose up --build

Frontend on http://localhost:5173, backend on http://localhost:8000. Both containers run production servers (gunicorn and nginx) as non-root users.

---

Configuration

Backend — backend/.env

Variable	Default	Description
DJANGO_SECRET_KEY	(dev fallback)	Django signing key. Required when DEBUG=False — startup fails if unset.
DEBUG	False	Enables the insecure dev fallback for the secret key. Never enable in production.
ALLOWED_HOSTS	localhost,127.0.0.1,[::1]	Comma-separated. * is rejected when DEBUG=False.
CSRF_TRUSTED_ORIGINS	(empty)	Comma-separated absolute origins.
CORS_ALLOWED_ORIGINS	http://localhost:5173,http://127.0.0.1:5173	Origins allowed to call the API from a browser. Never a wildcard.
LOG_LEVEL	INFO	Root logger level.
SECURE_SSL_REDIRECT	True	Applied only when DEBUG=False.
SECURE_HSTS_SECONDS	31536000	Applied only when DEBUG=False. 0 disables HSTS.
OPENAI_API_KEY	(empty)	Required for generation. Without it the API returns 503.
OPENAI_CHAT_MODEL	gpt-4o-mini	Model used for all text generation.
OPENAI_IMAGE_MODEL	dall-e-2	Model used for the cover image.
OPENAI_IMAGE_SIZE	512x512	Cover image dimensions.
OPENAI_TIMEOUT_SECONDS	60	Per-request timeout for the OpenAI client.
DB_ENGINE	django.db.backends.sqlite3	Set to django.db.backends.mysql or ...postgresql to use a server.
DB_NAME	<BASE_DIR>/db.sqlite3	Database name, or SQLite file path.
DB_USER / DB_PASSWORD / DB_HOST / DB_PORT	(empty)	Ignored by SQLite.

Non-SQLite engines need their driver installed separately — pip install mysqlclient or pip install "psycopg[binary]".

Frontend — frontend/.env

Variable	Default	Description
VITE_API_URL	http://localhost:8000/api	Base URL of the API, including the /api prefix. Inlined at build time, so it is public — never put a secret in a VITE_ variable.

---

API reference

All endpoints accept and return JSON. There is no authentication; deploy behind your own gateway if you need it.

Method	Endpoint	Description
GET	/healthz/	Liveness probe. Returns {"status": "ok"}.
POST	/api/search_topics/	Suggest article topics for a keyword.
POST	/api/select_topics/	List the sections available for a chosen topic.
POST	/api/generate_article/	Generate the full article.
—	/admin/	Django admin. No app models are registered.

POST /api/search_topics/

// request
{ "keyword": "kubernetes" }

// 200
{
  "keyword": "kubernetes",
  "topics": { "1": "Kubernetes For Beginners", "2": "Scaling Workloads", "3": "kubernetes" }
}

The keyword itself is always appended as the final topic so it can be used verbatim.

POST /api/select_topics/

Send either topic_id (with the topics map from the previous call) or custom_topic — never both.

// request
{ "topic_id": 1, "topics": { "1": "Kubernetes For Beginners" } }

// 200
{
  "selected_topics": { "1": "Kubernetes For Beginners" },
  "topic_sections": {
    "1": { "sections": { "1": "What Is A Pod", "2": "Services And Ingress" }, "type": "predefined" }
  }
}

For a custom topic the key is custom_1 and type is "custom".

POST /api/generate_article/

// request
{
  "selected_sections": [{ "section_name": "What Is A Pod" }],
  "selected_topics": { "1": "Kubernetes For Beginners" },
  "topic_sections": { "1": { "sections": ["What Is A Pod", "Services And Ingress"] } },
  "keyword": "kubernetes"
}

// 200
{
  "keyword": "kubernetes",
  "main_topic": "kubernetes",
  "selected_topic": { "1": "Kubernetes For Beginners" },
  "section_headers": ["What Is A Pod"],
  "introduction": "...",
  "conclusion": "...",
  "image_url": "https://...",
  "section_content": { "What Is A Pod": "..." }
}

Sections that the chosen topic did not offer are dropped; if none remain the request is rejected with 400.

Error responses

Status	Meaning
400	Validation failed, or the selection was inconsistent. Body carries the field errors.
502	The OpenAI API failed or returned unusable content.
503	OPENAI_API_KEY is not configured on the server.

A failed generation is never returned as a 200 with empty content.

---

Limitations

• No authentication and no rate limiting. Every request spends OpenAI credits, so do not expose this to the open internet without a gateway that enforces both.
• Cost and latency scale with section count. One article is 3 + n chat completions plus one image; a 6-section article takes roughly 30–60 seconds.
• Article state travels in the URL between pages. Very long topic lists can approach browser query-string limits.
• image_url points at OpenAI's CDN and expires (typically within an hour). Download it if you need to keep it.
• Generated content is unverified. The model can and does state things that are wrong. Treat every article as a first draft.
• No streaming. The full article is returned in one response.
• English only — prompts are hardcoded in English.

---

Development

Setup, the exact commands CI runs, and the project layout are documented in CONTRIBUTING.md. Security reporting and a pre-deployment checklist are in SECURITY.md.

Licence

Copyright (C) 2026 Article Generator contributors.

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version. It is distributed WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the LICENSE file for the full text.