The AI-Powered Misinformation Verification and Fact-Checking System is an intelligent, multimodal platform designed to detect, verify, and debunk false claims circulating online using advanced natural language processing, computer vision, and audio analysis models. It integrates multiple data input methods, including a Sidebar Upload Section for uploading documents (TXT, PDF, DOCX), images (JPG, PNG), and audio (MP3, WAV), a Fact-Check Database for manually adding verified claims with verdicts, explanations, and sources, and a Preloaded Data Module that auto-imports existing evidence and fact-checks from local directories at startup. The core functionality lies in its Claim Verification Tab, where a user can paste any suspicious claim; the system then performs a semantic search across the fact-check and evidence database using embeddings from

sentence-transformers/all-MiniLM-L6-v2

stored in ChromaDB, a persistent vector database that supports cosine similarity matching. Once the most relevant entries are retrieved, they are analyzed through Grok AI, which provides a structured output containing the verdict, confidence level, key findings, red flags, and evidence-based reasoning. In the Evidence Processing Module, text documents undergo extraction and embedding, images are processed through Grok Vision to detect manipulation, text overlays, and context cues, and audio files are transcribed via Whisper AI for voice deepfake or misinformation detection. The system’s Smart Search Engine uses vector similarity to find semantically similar claims, even if worded differently, ensuring high recall and accuracy. The Monitoring Mode continuously observes new misinformation trends, identifying manipulation tactics, analyzing spread potential, and assigning urgency levels based on risk factors. Furthermore, the Public Communication Generator simplifies technical fact-checking outputs into clear, 5th-grade reading-level public announcements with sections like “Fact,” “Fiction,” and “What to Do,” promoting transparency and public trust. Real-time Database Statistics and Verification History features provide users with insights into stored fact-checks, documents, and media while maintaining a searchable archive of all verification sessions with timestamps. Technically, the architecture follows a layered pipeline starting with User Input, passing through a Preprocessing Layer that extracts text, analyzes images, and transcribes audio, followed by an Embedding Layer that converts data into 384-dimensional vectors. These vectors are stored in the Vector Database (ChromaDB), categorized into Fact-Checks and Evidence Collections. The Retrieval Layer performs semantic search and returns the top relevant results to Grok AI, which carries out multi-stage analysis—claim verification, misinformation pattern detection, confidence scoring, and public message generation. The Structured Output Layer finally compiles the verdict, supporting evidence, and actionable recommendations for end users. For example, if a WhatsApp screenshot with a false government cash ban message is uploaded, Grok Vision extracts visible text, correlates it with related fact-checks, and determines that the claim is false with a 92% confidence score. The output cites official sources, highlights misleading phrasing, and generates a ready-to-share public update clarifying the truth. All embeddings and metadata are stored persistently under

misinfo_rag_db/

for long-term reliability, while session states (like cached models and history) are temporarily managed via

st.session_state

. This system stands out for being multimodal, AI-driven, and explainable, combining semantic understanding, vector search, and human-readable reporting. It’s scalable—capable of continuously expanding with new fact-checks and evidence—while maintaining transparency through structured reasoning and credible source references. Overall, it provides an end-to-end solution for combating misinformation: ingesting diverse media, semantically linking data, performing intelligent verification, and translating technical insights into public-friendly clarity, ultimately contributing to a more informed and resilient digital ecosystem.