dineshsrivasthav/adaptive_meta_learning_with_multi_agent_framework — reverse-engineered prompt

Build me a Python research pipeline for robust deepfake detection based on adaptive meta learning and a multi agent workflow. I want it to take local papers or PDFs plus optional web crawling, build a searchable knowledge base about new deepfake attack trends, then use several agents to turn that knowledge into useful prompts for creating new deepfake style training samples.

The app should be able to generate few shot image samples from base face images using Grounding DINO, SAM, and Stable Diffusion, then train a detector with Reptile style meta learning using normal images, fake images, generated samples, adversarial samples, and augmentations. Make it possible to run everything end to end from one main script, but also let me run the RAG, agents, sample generation, and training steps separately.

Please include a clear config file, command line options, environment variable handling for API keys, saved outputs, model checkpoints, and simple result metrics or plots. Look up current docs online if you need to.