awesome-generative-ai

Organizes curated Generative AI resources into a multi-level taxonomy (text generation, image generation, audio/speech/video, multimodal, code generation, etc.) with reverse chronological ordering and bidirectional linking. Uses a README.md-centric architecture where the main content file serves as the single source of truth, with auxiliary files (ARCHIVE.md, CITATION.bib, contributing.md) providing supplementary context and metadata. Resources are tagged with multiple dimensions (modality, tool type, capability) enabling cross-cutting discovery patterns.

Curates and organizes resources across the text generation modality, including Large Language Models (LLMs), prompt engineering techniques, Retrieval-Augmented Generation (RAG) systems, and LLM agents. Structures resources into subcategories covering model architectures (GPT, BERT, LLaMA variants), fine-tuning approaches, in-context learning, and agent frameworks. Maintains links to foundational papers, implementation guides, and production tools, with emphasis on reverse chronological ordering to surface recent advances in transformer architectures and instruction-tuning methods.

Aggregates resources for code generation and AI-assisted software development, including code completion tools (GitHub Copilot, Tabnine), code generation models (Codex, CodeLlama), and code-specific LLM applications. Organizes resources by capability (code completion, generation, refactoring, testing, documentation) and programming language support. Includes links to foundational papers, implementation frameworks, and production tools. Maintains reverse chronological ordering to surface recent advances in code understanding and generation.

Curates resources for datasets and benchmarks used in generative AI research and development, including training datasets (Common Crawl, LAION, The Pile), evaluation benchmarks (MMLU, HumanEval, COCO), and domain-specific datasets. Organizes resources by modality (text, image, audio, video, multimodal) and use case (pretraining, fine-tuning, evaluation). Includes links to dataset repositories, benchmark leaderboards, and papers describing dataset construction and evaluation methodologies. Maintains reverse chronological ordering to surface recent datasets and benchmarks.

Aggregates image generation resources organized into three primary subcategories: Stable Diffusion (open-source diffusion models and fine-tuning approaches), Advanced Image Generation Techniques (ControlNet, LoRA, inpainting, style transfer), and Image Enhancement (upscaling, restoration, quality improvement). Resources include links to model checkpoints, implementation frameworks (Diffusers, ComfyUI), research papers on diffusion processes, and community-built tools. Maintains chronological ordering of new techniques and model releases to surface recent advances in conditional generation and multi-modal control.

Organizes audio, speech, and video generation resources into three subcategories: Audio and Music Generation (text-to-music, music style transfer, sound synthesis), Speech Processing (text-to-speech, voice cloning, speech enhancement), and Video Generation (text-to-video, video synthesis, motion control). Curates links to foundational models (Jukebox, Bark, Stable Video Diffusion), implementation frameworks, and research papers. Resources are tagged by modality and capability, with reverse chronological ordering to surface recent advances in multimodal generation and temporal consistency.

Aggregates resources for multimodal models (vision-language models like CLIP, GPT-4V, LLaVA) and specialized applications (AI in games, code generation). Organizes resources by application domain rather than modality, reflecting the shift toward unified models that operate across text, image, audio, and video. Includes links to foundational papers, implementation frameworks, and domain-specific tools. Maintains reverse chronological ordering to surface recent advances in model scaling and cross-modal reasoning.

Implements a structured contribution process with formal guidelines (contributing.md), code of conduct (code-of-conduct.md), and citation metadata (CITATION.bib). Uses GitHub's pull request mechanism as the primary contribution channel, with community review and maintainer approval required before merging. Maintains auxiliary files for archived resources (ARCHIVE.md) and supporting information (AUXILIAR.md), enabling transparent version control and historical tracking of resource additions/removals. Reverse chronological ordering within categories ensures new contributions are immediately visible.

Implements a reverse chronological ordering strategy within each category and subcategory, placing newest resources at the top to surface recent advances and emerging tools. This architectural choice prioritizes temporal relevance over foundational importance, enabling rapid discovery of latest models, papers, and techniques. Supported by Git commit timestamps and contributor metadata, enabling tracking of when resources were added. Works in conjunction with the contribution workflow to automatically surface new additions without requiring manual curation or ranking.

Curates resources specifically focused on prompt engineering methodologies, including chain-of-thought prompting, few-shot learning, instruction tuning, and domain-specific prompt patterns. Organizes resources into guides, papers, and tools that teach prompt optimization techniques. Includes links to prompt engineering frameworks, best practices documentation, and case studies demonstrating prompt effectiveness across different model architectures (GPT, Claude, Llama). Maintains reverse chronological ordering to surface recent advances in prompt design and in-context learning.

Aggregates resources for Retrieval-Augmented Generation (RAG) systems, including vector databases, embedding models, retrieval frameworks, and hybrid search techniques. Organizes resources into categories covering semantic search, dense retrieval, sparse retrieval, reranking, and knowledge base construction. Includes links to foundational papers (e.g., Lewis et al. RAG), implementation frameworks (LangChain, LlamaIndex), and production tools (Pinecone, Weaviate, Milvus). Maintains reverse chronological ordering to surface recent advances in retrieval efficiency and multi-hop reasoning.

Curates resources for LLM agents and multi-agent systems, including agent frameworks (AutoGPT, LangChain agents, CrewAI), agent architectures (ReAct, tool-use patterns, memory systems), and multi-agent orchestration. Organizes resources by agent capability (planning, reasoning, tool use, memory management) and framework type. Includes links to foundational papers (e.g., ReAct), implementation frameworks, and production tools. Maintains reverse chronological ordering to surface recent advances in agent autonomy and multi-agent coordination.