Veracity Engine: AI-Powered Fact-Checker
An advanced RAG pipeline that verifies claims against real-time web evidence, analyzes source credibility, and ensures response reliability with self-consistency checks.
Project Overview
The Veracity Engine is a sophisticated tool that goes beyond a simple "true/false" verdict. When a user submits a claim via voice or text, the system performs real-time web searches using the Tavily API, vets the credibility of the sources, and extracts relevant evidence. It then employs advanced AI patterns like Chain-of-Thought and Self-Consistency with the Gemini LLM to generate a nuanced, reliable verdict. A custom evaluation system assesses content quality, evidence grounding, consistency, and relevance, with full source transparency.
Veracity Engine
Evidence-Based Verification
Key Features
🗣️ Multimodal Input
Accepts user claims through both typed text and high-quality spoken audio for maximum accessibility.
🌐 Live Web Search
Utilizes the Tavily Search API to gather the latest, most relevant information from across the web.
🛡️ Source Vetting
Automatically prioritizes information from a predefined list of reputable domains to improve reliability.
⚙️ Self-Consistency
Queries the LLM multiple times in parallel and selects the majority verdict to reduce errors and hallucinations.
🧪 Custom Evaluation
Evaluates responses for content quality, evidence grounding, consistency, and relevance using a custom LLM-based scoring system.
⚡ FAISS Caching
Uses a high-speed, in-memory FAISS vector store to cache claims and accelerate future checks of similar topics.
Fact-Checking Pipeline
Click on each component of the pipeline to learn more about its role in the system.
User Claim
Voice/Text Input
Web Retrieval
Tavily Search API
Evidence Synthesis
Gemini LLM
Nuanced Verdict
Output with Sources
Select a component to see details.
Skills Demonstrated
- ✓Grounded RAG Pipelines
- ✓Advanced Prompt Engineering (CoT, Self-Consistency)
- ✓Real-Time API Integration (Tavily, Gemini)
- ✓Custom LLM-based Evaluation
- ✓High-Performance In-Memory Caching (FAISS)
- ✓Multimodal I/O (Speech-to-Text, Text-to-Speech)