How aryamundra/veriwise works
Project type and purpose
• VeriWise is a full-stack web application for detecting manipulated media (DeepFakes) and verifying textual claims.
• The README confirms it uses Next.js for the frontend and FastAPI-based Python services for backend AI modules (DeepFake detection, fact-checking, multimodal preprocessing).
File structure
• Top level splits cleanly into frontend/ and backend/.
• Frontend: Next.js app under frontend/app with React components (.jsx), global styles, config files (next.config.mjs, jsconfig.json).
• Backend: Python module with:
• main.py – FastAPI entrypoint.
• AI service scripts: AI_Image.py, Deep_video.py, Manipulated.py, biasness.py, fact_verify.py.
• factcheck/ package: core submodules (CheckWorthy.py, ClaimVerify.py, Decompose.py, QueryGenerator.py, Retriever/serper_retriever.py) and utilities (api_config.py, data_class.py, multimodal.py, llmclient/, prompt/, etc.).
• requirements.txt, a log directory, test data under DeepFake/Test_data/.
README for additional insights
• Architecture block shows: Browser → Next.js → Backend API → DeepFake Service / Fact-Check Service → Multimodal Utils.
• Technology stack: Frontend (Next.js, Tailwind), Backend (FastAPI, Python), AI libs (OpenCV, PyTorch, Transformers), retrieval services (Gemini, Serper).