Fake News Detector AI — Streamlit + Gemini

This is a Streamlit web application utilizing Google's Gemini API to perform fact-checking on news stories, social media posts, and assertions. It separates factual claims from a document, verifies them with Gemini, and receives back a structured judgment with confidence values, justification, and evidence citations.

The goal is to provide a local, rapid prototype of an AI-driven fake news classifier with a simple, user-friendly UI.

---

✨ Features

• Interactive Streamlit UI – paste an article or claim and see immediate results.
• Claim extraction – automatically finds up to N claims from the input text.
• Structured analysis – all claims are labeled as REAL / FAKE / UNVERIFIED.
• Confidence scoring – numeric confidence (0–100) for every claim and overall verdict.
• Evidence citations – Gemini returns URLs/snippets if available (never hallucinating, if it is uncertain, it returns no sources).
• JSON Output – raw structured JSON is output for auditing or reuse.
• Flexible controls – choose model (e.g., gemini-2.5-flash), alter max claims, and set a thinking budget.
• Fully customizable UI - any colors can be chosen for the the theme of the web app.
• Screencast functionality - you can record and download your screen within the web app. (your data is only stored locally)

---

Getting Started

Prerequisites

• Python 3.9+
• A Gemini API Key from Google AI Studio (It's free)
• A SerpAPI Key from SerpAPI Dashboard (250 searches/month for free)

Installation

# Clone repo (if using git)
git clone https://github.com/CyberZenithX/Gemni-AI-Fact-checker.git
cd Gemni-AI-Fact-checker

Install dependencies

pip install streamlit google-genai pydantic serpapi beautifulsoup4 requests

Environment Setup

Set your Gemini API key as an environment variable (Optional):

export GEMINI_API_KEY="your_api_key_here"
export SERPAPI_API_KEY="your_serpapi_key"

Or, copy and paste the key directly into the Streamlit sidebar when running.

Run the App

streamlit run AI-fact-checker.py

---

Usage

1. Launch the app in your browser (Streamlit will serve a local URL).
2. Copy and paste a news article, tweet, or claim.
3. Click Run fact-check.
4. Check it out:

• Overall decision (REAL / FAKE / LIKELY REAL / LIKELY FAKE / UNVERIFIED)
• Decision at every level of claim with justification & evidence
• Unprocessed JSON output

---

⚠️ Limitations & Caveats

• The tool may not always have access to real-time web. Temporally sensitive or obscure claims may be flagged UNVERIFIED. (This depends on SerpAPI)
• Don't rely exclusively on AI decisions when making critical decisions. Always cross-check with independent fact-checking agencies.
• URLs provided can be partial when Gemini cannot find credible evidence. (Under further development)

---

Future Roadmap

• Add search engine cross-referencing (e.g., SerpAPI or NewsGuard) for better sourcing.
• Integrate Google Fact Check Tools API for stronger proof validation.
• Save results to a database for auditing, analysis and machine learning.
• Provide export capabilities (CSV / PDF report).

---

License

This project is provided as-is under the MIT License.

---

Acknowledgments

• Google Gemini API
• SerpAPI
• Streamlit
• Fact-checking organizations like PolitiFact, Snopes, and Full Fact, whose practices inspired the structure of this tool.

---

Summary: This project is a prototype AI fact-checking assistant that combines Gemini’s reasoning and SerpAPI's searching power with a simple Streamlit UI to provide quick and interpretable fake news detection.