Google ADK Deep-Dive: Agentic RAG, SQL + Live Web Search in 1 Hour | Live Build | AI Bros Pod EP 32
In Episode 32, Nisaar and Rohan go end-to-end with Google ADK to build four working agents—a Base Agent, an Agentic RAG agent powered by Pinecone, a SQL Agent over SQLite, and a Web Search Agent grounded with Google Search. We start from a blank repo, set up the environment, scaffold agents, wire up tools, and test everything live (including traces, token usage, and a web UI). We also use real Tesla earnings data (PDF + SQL) to stress-test retrieval, trend questions, and multi-table SQL.
Timestamps
00:00 Stream Start & Introduction (Claude 4.5 context)
02:00 Agenda: Base, Agentic RAG, SQL, Web Search (Tesla PDF + DB)
04:00 Initial Setup: venv + pip install google-adk
06:00 Creating the Base Agent (Gemini, name/desc/instructions, API key)
09:30 Running the Base Agent + Web UI demo on localhost:8000
12:00 Debugging the Dev Panel: States, Sessions, Events, Trace Graph
15:00 Agentic RAG: Connecting to Pinecone (Tesla earnings vectors)
17:30 RAG Tools: Embeddings + pinecone_retrieval (top-5 chunks)
23:00 RAG Test: Revenue question + page citations (p. 3–5)
26:00 RAG Stress Test: Diner features, Model 3/Y trends & caveats
36:00 SQL Agent Architecture: list → describe → sample → generate → run
40:00 SQL Agent Tools: SQL_list_table, SQL_describe_table, SQL_run_SQL
44:00 Prompting the SQL Agent to be safe & deterministic
46:30 SQL Tests: SUM cars sold, pick correct table, pricing queries
52:00 Multi-Table SQL: pricing ↔ deliveries join scenario
55:00 Web Agent Overview: grounding & source citing
57:30 Web Agent Setup: adk create web_agent, register GoogleSearch
59:30 Web Agent Live Test: “Why is gold price rising?” + citations
YouTube Tags (copy all):
Google ADK, ADK agents, agentic RAG, RAG pipeline, Pinecone, Pinecone vector database, Gemini, Google AI, LLM agents, SQL agent, SQLite, database agents, Google Search Grounding, web search agent, retrieval augmented generation, vector search, embeddings, OpenAI embeddings, Tesla earnings, Tesla data, AI coding tutorial, AI agents tutorial, build AI agents, developer tools, trace graph, prompt engineering, tool use, LangChain alternative, Python AI, VS Code, autonomous agents, enterprise AI, production AI, AI workflows, AIBROS podcast, Nisaar, Rohan, Claude 4.5, ChatGPT, Open Source AI, Generative AI, Large Language Models, machine learning
Hashtags:
#AIBROS #GoogleADK #AIagents #RAG #Pinecone #Gemini #LLM #SQL #VectorDB #Python #MachineLearning #AI #Developers #GoogleAI
Tags (paste into YouTube “Tags”)
AIBros Podcast, Google ADK, Agent Developer Kit, ADK agents, agentic RAG, RAG systems, retrieval augmented generation, SQL agent, database agent, web agent, browsing agent, toolformer, tool use, agent orchestration, AI agents, autonomous agents, LLM tools, LLM evaluation, prompt engineering, vector databases, embeddings, grounding, AI observability, guardrails, safety, hallucination reduction, production AI, Python AI, VS Code AI, data engineering, analytics, enterprise AI, generative AI, machine learning, large language models, open source AI, AI livestream, coding live, tech breakdown, AI tutorial, Google AI
Hashtags
#AIBros #GoogleADK #AgentDeveloperKit #AIagents #AgenticRAG #SQLAgent #WebAgent #RAG #GenerativeAI #LLMs #MachineLearning #OpenSource #Python #AIEngineering #LiveStream #TechTutorial