With the rapid spread of digital misinformation, journalists, researchers, and general users require transparent, source-backed claim evaluation. FactGuard AI solves this by executing automated web searches and generating transparent, audit-ready reports.
Designed an automated search-retrieval pipeline that ranks source relevance and credibility. Claim decomposition, credibility evaluation, and evidence extraction are offloaded to Llama 3.1 8B running locally on consumer hardware via Ollama.
Built a high-performance frontend with React, Vite, and TanStack Query. Integrated a Supabase Edge Function to choreograph Deno-based search calls, nomic-embed-text deduplication, and Llama 3.1 reasoning APIs. Fully secured using Postgres Row-Level Security (RLS).
User Submit → Dashboard Page → Supabase Edge Function → Web Search API → Source Evaluation (Ollama embeddings) → Evidence Analysis (Llama 3.1 8B) → RLS Database Write → React Dashboard Report.
React
Modern frontend framework with a fully responsive, dark-mode native interface
TypeScript
Ensures type-safe client data and reliable document structures
Ollama
Hosts Llama 3.1 8B and nomic-embed-text locally on consumer hardware
Llama 3.1 8B
Executes reasoning, claim decomposition, source ranking, and verdict synthesis
Supabase
Postgres database, Row-Level Security, email/Google authentication, and Edge Functions
nomic-embed-text
Generates high-quality text embeddings used for source de-duplication
Tailwind CSS
Sleek dark-mode native interface utilizing semantic HSL design tokens
