This is a personal learning path repository for learning to build agents in python.
NOTE: This is not meant to teach, but to learn.
The learning path is divided into 3 Tiers:
An agent that can use Python functions as tools — a calculator, a weather fetcher, a file reader. This is the "Hello World" of agents. Learn the ReAct loop (Reason → Act → Observe) which is the heartbeat of every agent.
An agent that remembers things across conversations using a simple vector store (like ChromaDB). Understand the difference between short-term context and long-term memory — a key design decision in every real agent.
Give the agent a high-level goal (e.g., "Research Python best practices and write a summary") and it breaks it into subtasks, executes them sequentially, and reports back. Core concept: planning and self-prompting.
An agent that takes a question, searches the web (via Tavily/SerpAPI), scrapes relevant pages, synthesizes an answer, and cites sources. Combines tool use, chaining, and output formatting.
Point the agent at a local repo and ask it questions: "Find all functions that handle auth", "Refactor this module". You'll learn RAG + agents together with file I/O tools.
Connect to Gmail and Google Calendar APIs. The agent reads emails, drafts replies, schedules meetings, and handles follow-ups. This is where real-world reliability and error handling become the main lesson.
Two agents with distinct roles: a Researcher agent that gathers data, and a Writer agent that turns it into a blog post or report. They communicate via a shared message bus. This teaches agent orchestration.
Give a spec, and a team of agents (Planner → Coder → Tester → Debugger) collaboratively write, run, test, and fix code. This is the capstone — it ties together everything from all tiers.
| Layer | Recommended | Used |
|---|---|---|
| LLM API | Anthropic (claude-sonnet) or OpenAI | Ollama + qwen2.5:7B |
| Agent Framework | Raw Python first, then LangGraph | Raw Python first, then LangGraph |
| Memory | ChromaDB or FAISS | ChromaDB |
| Tool Execution | Function calling / MCP | Function calling |
| Orchestration | LangGraph or AutoGen (Tier 3) | — |