🧠 Ollama4Truth

Ollama4Truth is an open-source framework for misinformation detection and fact-checking using Large Language Models (LLMs) through the Ollama ecosystem.

This project aims to explore, evaluate, and democratize the use of open-source LLMs for identifying and mitigating online disinformation — particularly in Portuguese (PT-BR) and English contexts.

---

🎯 Objectives

• Develop a modular and reproducible pipeline for fact-checking and misinformation identification.
• Evaluate open LLMs’ capabilities in detecting false or misleading content.
• Foster open collaboration and transparent evaluation within the AI research community.

---

⚙️ Configuration

1. Install Ollama and download any model:

curl -fsSL https://ollama.com/install.sh | sh
ollama run gemma3:1b

2. Install requirements (with Python 3.10):

pip install -r requirements.txt

3. Create a .env file:

OLLAMA_MODEL=gemma3:1b
GOOGLE_API_KEY=YOUR_GOOGLE_API_KEY
GOOGLE_CSE_ID=YOUR_CSE_ID

🚀 How to run?

1. Start the server:

uvicorn api:app --reload

---

2. Send a POST request to http://localhost:8000/analyze with the claim:

{
  "claim": "O café ajuda a melhorar a memória de longo prazo."
}

🖥️ Technologies

• 🦙 Ollama — local LLM inference
• 🔍 Google Search API — open evidence retrieval
• 🤗 Transformers — tokenization and model loading
• 🧮 PyTorch — inference backend
• 📄 BM25 / FAISS — ranking and document retrieval
• 🧰 Python (3.10+)

---

🪶 License

Released under the MIT License — free for research and open-source use.

---