TruthLens 🔍

AI Hallucination Detector — Gemma 4 Good Hackathon 2026

TruthLens is an open-source web application that detects hallucinations in AI-generated text. Paste any AI-generated content into the app, and TruthLens uses Gemma 3 4B running locally via Ollama to evaluate every sentence for accuracy, uncertainty, and hallucinations — with zero data leaving your machine.

Built for the Safety & Trust track of the Gemma 4 Good Hackathon on Kaggle.

Live Demo Link: Demo

---

✨ Features

• 🔍 Sentence-level analysis — every sentence is individually classified
• 📊 Trust Score (0–100) — instant overall reliability rating
• 🟢🟡🔴 Color-coded results — Accurate / Uncertain / Hallucination
• 💡 Hover tooltips — plain-language explanation for every sentence
• 🔒 100% local — your text never leaves your machine
• ⚡ Edge-optimized — runs on consumer hardware, no GPU required

---

🛠️ Tech Stack

Layer	Technology
Frontend	React 18 (Create React App) + plain CSS
Backend	FastAPI (Python)
AI Model	Gemma 3 4B via Ollama (edge deployment)
Model Target	Gemma 4 27B (high-accuracy environments)
Communication	REST API (HTTP/JSON)

---

📁 Project Structure

TruthLens/
├── backend/
│   ├── main.py          # FastAPI app + endpoints
│   ├── analyzer.py      # Gemma hallucination analysis logic
│   ├── requirements.txt
│   └── .venv/           # Python virtual environment
├── frontend/
│   ├── src/
│   │   ├── App.js
│   │   ├── index.css
│   │   └── components/
│   │       ├── TrustScore.jsx
│   │       ├── HighlightedText.jsx
│   │       └── ResultPanel.jsx
│   ├── package.json
│   └── public/
└── README.md

---

🚀 Setup Instructions

Prerequisites

• Node.js (v18 or higher)
• Python (v3.10 or higher)
• Ollama

---

Step 1 — Install Ollama

Download and install Ollama for your OS from: 👉 https://ollama.com/download

Verify installation:

ollama --version

---

Step 2 — Pull the Gemma model

# Recommended — fast, edge-optimized (3GB)
ollama pull gemma3:4b

# Optional — highest accuracy (9.6GB, needs 16GB+ RAM)
ollama pull gemma4

---

Step 3 — Verify Ollama is running

Ollama starts automatically on Windows. Verify with:

ollama list

You should see gemma3:4b in the list. If Ollama isn't running:

ollama serve

---

Step 4 — Start the Backend

Open a terminal in the backend folder:

Windows (PowerShell):

cd backend
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
uvicorn main:app --reload

macOS / Linux:

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload

Backend runs at: http://localhost:8000

Verify it's working:

http://localhost:8000/
→ {"status": "TruthLens backend is running!"}

http://localhost:8000/test
→ Returns a sample analysis (tests Gemma connection)

---

Step 5 — Start the Frontend

Open a second terminal in the frontend folder:

cd frontend
npm install
npm start

App opens automatically at: http://localhost:3000

---

🔌 API Reference

POST /analyze

Analyzes AI-generated text for hallucinations.

Request:

{
  "text": "Your AI-generated text here"
}

Response:

{
  "trust_score": 65,
  "overall_verdict": "Contains some accurate facts but includes a clear hallucination.",
  "sentences": [
    {
      "text": "Einstein was born in Germany.",
      "risk_level": "green",
      "explanation": "Accurate — Einstein was born in Ulm, Germany in 1879."
    },
    {
      "text": "He invented the television.",
      "risk_level": "red",
      "explanation": "False — The television was invented by Philo Farnsworth, not Einstein."
    }
  ]
}

Risk levels:

Value	Meaning
green	Accurate and verifiable
yellow	Uncertain or unverifiable
red	Hallucination or clearly false

---

🧪 Example Test Inputs

Test 1 — Einstein (clear hallucination):

Albert Einstein was born in Ulm, Germany in 1879.
He won the Nobel Prize in Physics in 1921.
Einstein attended Harvard University for his PhD.
He also invented the telephone.

Test 2 — Medical (high stakes):

The human body contains 206 bones in adults.
Penicillin was discovered by Alexander Fleming in 1928.
Drinking bleach in small amounts can cure bacterial infections.
The human brain uses approximately 20% of the body's total energy.

---

⚙️ Configuration

Switching models

In backend/analyzer.py, change the model name:

# Fast, edge-optimized (recommended for demos)
"model": "gemma3:4b"

# Highest accuracy (requires 16GB+ RAM, slower)
"model": "gemma4"

Timeout

For slower machines or larger models, increase the timeout:

timeout=300  # seconds

---

🏗️ How It Works

User pastes text
      ↓
React frontend (App.js)
      ↓ POST /analyze
FastAPI backend (main.py)
      ↓
analyzer.py → Ollama API (localhost:11434)
      ↓
Gemma 3 4B processes prompt
      ↓
Returns structured JSON
      ↓
Frontend renders:
  • Trust Score gauge
  • Color-coded sentences
  • Hover tooltips with explanations

---

🐛 Troubleshooting

Problem	Fix
500 Internal Server Error	Check FastAPI terminal for details. Usually Ollama timeout — increase timeout=300 in analyzer.py
Could not connect to backend	Make sure FastAPI is running at localhost:8000
Ollama not responding	Run ollama list to verify. Restart with ollama serve
Results show 0 sentences	Open F12 → Console in browser and check for errors
First request very slow	Normal — model loads into RAM on first request. Subsequent requests are faster

---

📝 Notes

• CORS is enabled for all origins in development
• The /test endpoint at http://localhost:8000/test tests the Gemma connection directly
• Model stays loaded in RAM for 10 minutes between requests (keep_alive: 10m)
• For production, restrict CORS origins and add rate limiting

---

📄 License

MIT License — free to use, modify, and distribute.

---

🙏 Acknowledgements

• Google Gemma — open model family
• Ollama — local model runtime
• FastAPI — Python web framework
• Kaggle — Gemma 4 Good Hackathon platform

---

Submitted to the Gemma 4 Good Hackathon 2026 — Safety & Trust Track