4 Roadmaps33+ PapersRight Order & ContextFree to Read

ResearchRoadmaps

No paper exists in isolation. Read them in the right order, with full context between each one. RAG, Agents, RLHF, Multimodal: every major AI topic covered as a curated reading path.

Roadmaps

Papers

Coming Soon

Free

to Read

Start with RAG Evolution →Browse All Roadmaps ↓

0Active Roadmaps

0+Papers Covered

0Coming Soon

Freeto Read

4 Roadmaps. 33+ Papers. Full Context.

Each roadmap tells the story of a field, paper by paper, with the context between each one that textbooks and courses skip.

RAG8 papers

Retrieval-Augmented Generation

Lewis et al. (2020) → ColBERT → Self-RAG → RAPTOR → Corrective RAG → GraphRAG. The full evolution of retrieval-augmented generation, with production architecture at every step.

Dense RetrievalRerankingSelf-RAGGraphRAG

Read Roadmap →

Agents9 papers

AI Agents & Tool Use

Chain-of-Thought → ReAct → Toolformer → HuggingGPT → AutoGPT → LangGraph. From prompting tricks to production stateful multi-agent systems.

Chain-of-ThoughtReActTool UseMulti-Agent

Read Roadmap →

RLHF7 papers

Reinforcement Learning from Human Feedback

InstructGPT → Constitutional AI → DPO → ORPO → KTO. Alignment techniques from first principles, from reward modeling to training-free preference optimization.

InstructGPTDPOConstitutional AIReward Modeling

Read Roadmap →

Multimodal9 papers

Vision-Language Models

CLIP → Flamingo → LLaVA → GPT-4V → Gemini → CogVLM. How vision and language models unified into a single architecture, and what it means for production systems.

CLIPLLaVAVision-LanguageGPT-4V

Read Roadmap →

3 More Roadmaps Coming

Subscribe to AI Letters to be notified when new roadmaps go live.

DiffusionComing Soon

Diffusion Models for Generation

DDPM → DDIM → Stable Diffusion → SDXL → DiT. From the math of denoising to production image generation.

ReasoningComing Soon

LLM Reasoning & Planning

CoT → Tree-of-Thought → o1/DeepSeek-R1. How reasoning emerged in language models and what it costs.

MoEComing Soon

Mixture of Experts

GShard → Switch Transformer → GLaM → Mixtral. Why the future of LLMs is sparse activation.

Start with RAG Evolution

RAG is the most production-deployed AI technique of the past 3 years. The roadmap takes you from the original paper to GraphRAG, with the architecture implications of each step.

Read the RAG Roadmap →Back to Research Lab