newsREPORTERai 🚀📰

AI-driven news generation using Google Gemini + CrewAI agents. Surfaces emerging trends and turns them into insightful, reader-friendly articles.

Highlights

• Dual-Agent Pipeline: news_researcher + news_writer orchestrated by CrewAI.
• LLM: Google Gemini (Gemini 1.5 Flash) via langchain_google_genai.
• Search Tooling: Serper-powered search for web research.
• FastAPI Backend: Modular API with health/readiness and generation endpoints.
• Robust Fallback: If CrewAI stack isn’t available, the system uses Gemini directly to ensure real content is returned.

---

Table of Contents

• Getting Started (Conda-first)
• Project Structure
• Endpoints
• Testing
• Architecture
• Screenshots
• Troubleshooting & Notes
• Roadmap
• FAQs
• Contributing
• Security

Getting Started (Conda-first)

We recommend using Conda for a smooth macOS setup and a compatible onnxruntime build.

1) Clone

git clone https://github.com/austinLorenzMccoy/newREPORTERai.git
cd newREPORTERai

2) Environment variables

Create backend/.env (copy the example):

cp backend/.env.example backend/.env

Then populate at least:

• GOOGLE_API_KEY=<your_google_api_key>
• SERPER_API_KEY=<your_serper_api_key>

Optional service config:

• APP_NAME, APP_ENV, LOG_LEVEL, PORT, CORS_ORIGINS

3) Create the Conda environment

conda env create -f backend/environment.yml
conda activate newreporter_backend

4) (Optional) CrewAI-friendly pins

If you want a tighter dependency matrix for CrewAI + LangChain components:

pip install -r backend/requirements-crewai-full.txt

5) Run the API

uvicorn newreporter_backend.main:app \
  --app-dir backend \
  --host 127.0.0.1 \
  --port 8000 \
  --reload

Health checks:

curl http://127.0.0.1:8000/health
curl http://127.0.0.1:8000/ready

6) Try the demo client

python backend/api_demo.py --base-url http://127.0.0.1:8000 --self-check
python backend/api_demo.py --base-url http://127.0.0.1:8000 --generate --topic "AI in healthcare"

Behavior:

• If the full CrewAI pipeline is available, you’ll get CrewAI-authored content and an output_file path (e.g., new-blog-post.md).
• If CrewAI imports fail (e.g., memory stack issues), the fallback produces real content via Gemini (no output_file).

---

Project Structure

backend/
  environment.yml
  requirements-crewai-full.txt
  pyproject.toml
  setup.py
  requirements.txt
  pytest.ini
  README.md
  .env.example
  api_demo.py
  newreporter_backend/
    __init__.py
    config.py
    main.py
    api/
      __init__.py
      routes.py
      health.py
      schemas.py
      deps.py
    domain/
      agents.py
      tasks.py
      tools.py
    integrations/
      crew_runner.py
    services/
      __init__.py
      sample_service.py
  tests/
    __init__.py
    conftest.py
    test_health.py
    test_services_sample_service.py

new-blog-post.md        # Latest generated article (when using CrewAI with output)

---

Architecture

High-level flow:

[Client / api_demo.py]
        |
        v
 [FastAPI /api/v1/generate]  -->  [integrations/crew_runner.run_generation]
                                        |  (Primary) CrewAI: Agents + Tasks
                                        |       - domain/agents.py (memory disabled by default)
                                        |       - domain/tasks.py
                                        |       - domain/tools.py (Serper search)
                                        |  (Fallback) Direct Gemini generation
                                        v
                                 [Article Markdown]
                                        |
                                        v
                              new-blog-post.md (when CrewAI writes output)

• CrewAI path yields richer, web-researched content and writes to new-blog-post.md.
• Fallback guarantees real output even if memory stack or imports fail.

---

Screenshots

Add screenshots/GIFs of the API demo or rendered article here:

docs/
  screenshots/
    demo-call.png
    article-preview.png

---

Endpoints

• GET /health – liveness
• GET /ready – readiness
• GET /api/v1/ping – sample
• POST /api/v1/generate – triggers the CrewAI workflow or the Gemini fallback

Example:

curl -X POST \
  http://127.0.0.1:8000/api/v1/generate \
  -H 'Content-Type: application/json' \
  -d '{"topic": "AI in healthcare"}'

---

Testing

pytest backend/tests -q

---

Troubleshooting & Notes

• Conda + onnxruntime (macOS): The provided backend/environment.yml selects a compatible macOS build to avoid binary issues.
• CrewAI vs Fallback: The system always returns real content. When CrewAI is unavailable (e.g., memory deps), it falls back to direct Gemini generation.
• API Keys: Missing GOOGLE_API_KEY or SERPER_API_KEY will limit capabilities (no search or LLM). Ensure backend/.env is populated.
• Port conflicts: If :8000 is busy, start uvicorn on another port, e.g., --port 8010, and pass the same base URL to api_demo.py.
• Editable install (optional):

  pip install -e backend

---

Roadmap

• Richer prompt templates for different verticals (e.g., fintech, biotech, energy)
• Streaming responses over SSE/WebSocket
• Pluggable search providers (Google, Tavily, Bing)
• Dockerfile + CI workflow
• Vector memory re-introduction with stable dependency profile

---

FAQs

• Why do I sometimes see a “Placeholder Article”?
  • CrewAI executed but lacked live search context. Set SERPER_API_KEY in backend/.env and re-run.
• How do I avoid macOS onnxruntime issues?
  • Use the provided Conda env (backend/environment.yml) which selects a compatible build.
• Where is the article saved?
  • CrewAI path writes to new-blog-post.md; fallback returns content in the API response (no file).
• Can I run without Conda?
  • Yes, but Conda is recommended on macOS. A backend/requirements.txt exists if you prefer virtualenv.

Contributing

Contributions are welcome! Fork the repo, create a feature branch, and open a PR. Please keep secrets out of git and use .env locally.

---

Security

• Do not commit real API keys or secrets. Use .env locally and a secret manager in production.
• If sensitive keys were committed by mistake, rotate immediately and purge history if needed.

For more backend details, see backend/README.md.