ramirocerda

PROJECT / ai-agents-hub

AI Agents Hub — AI agent lab

Personal Active

Learning and reference monorepo for Google ADK and LangGraph: from small agents (PlanReAct, output_schema, MCP, code executor) to full-stack apps in production. Includes Customer Support Chat and Career Coach (Vertex AI Agent Engine + Cloud Run).

Problem

Learning agents for real means juggling frameworks, tools, and deploys. Without a structured monorepo you end up with loose notebooks, duplicated prompts, and no path to prod.

What I did

Built the hub in layers: ADK lab (YAML/Python tutors, PlanReAct via LiteLLM/OpenRouter, Pydantic product extractor, Google Search research agent, MCP file reader, BuiltInCodeExecutor math) + LangGraph notebook (email assistant) + two products. Customer Support Chat (React 19 + FastAPI + ADK). Career Coach with BuiltInPlanner, 3 deterministic tools, sessions, and a multi-stage Cloud Run/Vertex AI deploy with ADC. Ruff + pre-commit + ADK vs LangGraph architecture docs.

Highlights

  • Layered ADK lab: PlanReAct, Pydantic output_schema, an MCP file reader, and a code executor.
  • LangGraph notebook for an email assistant.
  • Full-stack Customer Support Chat (React 19 + FastAPI + ADK).
  • Career Coach with BuiltInPlanner and 3 deterministic tools, deployed to Cloud Run + Vertex AI Agent Engine via ADC (no API keys in prod).
  • Ruff + pre-commit and docs comparing ADK vs LangGraph.

Stack

  • Python 3.13
  • Google ADK
  • LangGraph
  • FastAPI
  • React 19
  • Vertex AI
  • Cloud Run
  • LiteLLM

Links