adongwanai/AgentGuide

What it solves

AgentGuide is an open-source knowledge base designed to bridge the gap between simple LLM API calls and the creation of reliable, production-ready AI Agent systems. It addresses common pain points such as agents failing during long tasks, lack of systematic architecture for multi-agent coordination, and the difficulty of translating technical projects into a competitive professional resume for AI roles.

How it works

The project organizes a comprehensive learning path divided into two primary tracks: Algorithm Engineering (focused on research, innovation, and experimental validation) and Development Engineering (focused on system architecture, reliability, and business deployment). It provides a structured curriculum covering:

• Agent Harnessing: Managing state, tool registration, permissions, and observability.
• Context Engineering: Handling memory, retrieval, and prompt caching.
• RAG Systems: Implementing hybrid retrieval, GraphRAG, and multimodal RAG.
• Evaluation & Safety: Building eval sets, trace/replay mechanisms, and sandbox environments.
• Post-training: Guidance on SFT, DPO/GRPO, and trajectory data synthesis.

Who it’s for

• Aspiring AI Agent, RAG, or LLM Application Engineers.
• Developers looking to transition into AI roles.
• Researchers focusing on Agentic RL, world models, or embodied intelligence.
• Job seekers needing a structured path from learning to securing an offer.

Highlights

• Job-Oriented Approach: Every knowledge point is mapped to how it is tested in interviews and how to describe it on a resume.
• Dual-Track Learning: Separate, specialized roadmaps for algorithm-focused and engineering-focused roles.
• Comprehensive Tech Stack: Covers modern frameworks like LangGraph, OpenAI Agents SDK, and protocols like MCP (Model Context Protocol).
• Practical Frameworks: Includes a "1-2-5 Job Search Framework" and a 5-step method for landing real-world projects.