The world's best open-source AI & LLM engineering resources — curated, annotated, and organized into one living repository.
A curated collection of the best GitHub repositories, papers, and resources for LLM engineering, RAG, AI agents, fine-tuning, prompt engineering, vector databases, and ML fundamentals — with original guides, decision frameworks, and learning paths built on top.
Stop searching. Start learning.
| # | Section | What It Covers |
|---|---|---|
| 01 | ML & DL Foundations | Math, algorithms, deep learning from scratch |
| 02 | Large Language Models | How LLMs work, training, inference, open-source models |
| 03 | Transformers & Architectures | Attention, BERT, GPT, ViT, Mamba, MoE — full lineage |
| 04 | RAG | Retrieval-Augmented Generation, production patterns |
| 05 | AI Agents | Agent design patterns, multi-agent systems, frameworks |
| 06 | Model Context Protocol | MCP spec, servers, SDKs, quick-start template |
| 07 | Vector Databases | Embeddings, FAISS, Qdrant, Chroma, Weaviate, Milvus |
| 08 | Fine-Tuning & PEFT | LoRA, QLoRA, DPO, RLHF, SFT — with decision guide |
| 09 | Prompt Engineering | CoT, few-shot, ReAct, libraries, pattern catalog |
| 10 | Tools & Infrastructure | LangChain, vLLM, Ollama, Dify, serving, observability |
| 11 | Coding & ML Interviews | DSA, ML system design, 30-day AI interview track |
| 12 | Data Science | Stats, SQL, EDA, feature engineering, daily practice |
| 13 | Must-Read Papers | Seminal research with TL;DRs and implementations |
| 14 | Learning Paths | Role-specific roadmaps: LLM engineer, researcher, DS |
| 15 | Newsletter Picks | Editorial intake queue — new additions land here first |
These are our proof of work — original guides built on top of the curated resources:
| Artifact | Section | What It Is |
|---|---|---|
| RAG Decision Matrix | 04-rag | Choose chunking strategy, retriever, and reranker by use case |
| Agent Pattern Playbook | 05-agents | 10 agent design patterns with code sketches |
| MCP Quick-Start Template | 06-mcp | Production-ready FastMCP server skeleton |
| Prompt Pattern Catalog | 09-prompts | Prompting patterns by use case with templates and failure modes |
| Vector DB Comparison | 07-vector-databases | Feature + latency comparison across Qdrant, Chroma, Weaviate, etc. |
| When to Fine-Tune | 08-fine-tuning | RAG vs. fine-tune vs. prompt engineering decision flowchart |
| AI Stack Guide | 10-tools-and-infra | Stack decisions by team size |
| 30-Day Interview Track | 11-coding-and-interviews | AI engineering interview prep plan |
| Paper TL;DR Index | 13-papers | 10 seminal papers, each with 3-sentence summary + implementation |
| Architecture Family Tree | 03-transformers | Model architecture lineage from 2017 to 2026 |
| Foundations Roadmap | 01-foundations | Beginner → Practitioner → Researcher learning path |
| State of LLMs | 02-llms | Monthly open-source model landscape snapshot |
Curation → Annotation → Synthesis. We never claim authorship of curated content. Every resource is linked to its original author with full attribution. Our proof of work is the layer we build on top: decision guides, learning paths, templates, and original artifacts that make the curated content immediately actionable.
The intake pipeline: New resources from our newsletter or community land in 15-newsletter-picks/ first. Each month, the best ones graduate to their permanent section. This keeps the repo living and tied to our editorial work.
"I want to build a RAG system" → 04-rag/ → RAG Decision Matrix → 07-vector-databases/
"I want to fine-tune an open-source LLM" → When to Fine-Tune → 08-fine-tuning/
"I want to build an AI agent" → Agent Pattern Playbook → 05-agents/ → 06-mcp/
"I want to prep for an AI engineering interview" → 30-Day Interview Track
"I'm new to ML/AI" → Foundations Roadmap → 14-learning-paths/
See CONTRIBUTING.md. You can add a new resource in under 5 minutes by opening a PR targeting 15-newsletter-picks/README.md.
All curated repos belong to their original authors. See CREDITS.md for the full attribution log. We link, layer, and synthesize — we never claim ownership of others' work.
MIT — see LICENSE. Original content (guides, templates, decision frameworks) is freely usable with attribution.
Updated monthly · v1.0 · June 2026