Matt Dancho's 90-Day AI Engineer Roadmap: 10 GitHub Repos Over Courses

On September 19, Matt Dancho, founder of Business Science, published a tweet thread outlining a 90-day path to becoming an AI engineer: skip paid courses and build projects directly with 10 GitHub repos, since these skills already appear heavily in job descriptions.

Confirmed

• Item 1 is LangChain, which Dancho says nearly every AI engineer role touches.
• Item 2, LangGraph, is a low-level orchestration framework for building long-running stateful agents — what JDs call "agentic workflows."
• Item 3, LlamaIndex (52.2k stars), is described as the de facto framework for RAG and document agents.
• Item 4, CrewAI (34.7k stars), is an open-source Python multi-agent framework organized around roles and tasks; Dancho says it is used in production by Fortune 500 companies.
• Items 5 and 6 are the Rust-based vector database Qdrant and the RAG evaluation framework Ragas, which quantifies hallucination, faithfulness and relevance, and can generate production-aligned test data when no test set exists.
• Item 7, Ollama (181k stars), runs open-source LLMs locally with one command; README shows support for Kimi, GLM and more, and JDs increasingly demand local inference experience for cost and privacy reasons.
• Item 8 is punkpeye's awesome-mcp-servers (95.3k stars), indexing production-ready MCP servers — Dancho calls MCP the newest skill appearing in job posts.
• Item 9 is Shubhamsaboo's awesome-llm-apps (139k+ stars, 20.4k forks), with 100+ open-source AI agents, Agent Skills and RAG apps under Apache-2.0, all runnable real code.

Why it matters

• The list turns the "projects over courses" philosophy into a concrete, actionable self-study roadmap for career switchers.
• Its selection criteria — appearing in JDs, de facto standards, Fortune 500 production use — reflect real hiring demand trends: orchestration frameworks, RAG, evaluation, local inference and the MCP ecosystem.
• High star counts across the repos (Ollama 181k, awesome-llm-apps 139k, awesome-mcp-servers 95.3k) corroborate the community momentum behind these directions.

2026-09-18 ~ 2026-09-19 · 11 related posts

• [source] awesome-llm-apps: 100+ open-source AI agents and skills with 139k GitHub stars — Saboo_Shubham_ · 2026-09-18
• [source] Want to become an AI engineer in 90 days? Build from these 10 GitHub repos, not courses — mdancho84 · 2026-09-19
• LangGraph ranked #2 on 10-repo AI engineer list, powering agents at Klarna and Replit — mdancho84 · 2026-09-19
• LlamaIndex ranked #3 on 10-repo AI engineer learning list with 52k GitHub stars — mdancho84 · 2026-09-19
• CrewAI lands at No.4 on AI engineer's 10-repo learning list, with 58.8k GitHub stars — mdancho84 · 2026-09-19
• CrewAI runs multi-agent teams in Fortune 500 production; Qdrant powers vector search in Rust — mdancho84 · 2026-09-19
• Ragas is the standard framework for measurable RAG evaluation: hallucination, faithfulness, relevancy — mdancho84 · 2026-09-19
• Ollama hits 181k stars: run open-source LLMs locally in one command — mdancho84 · 2026-09-19
• Awesome MCP Servers: a 95k-star index of every production MCP server — mdancho84 · 2026-09-19
• Awesome LLM Apps ships 100+ open-source RAG and agent templates with 139k GitHub stars — mdancho84 · 2026-09-19

1 near-duplicate retellings: mdancho84