Upgrade to Pro — share decks privately, control downloads, hide ads and more …

adk-samples に学ぶデータ分析 LLM エージェント開発

Avatar for na0 na0
November 23, 2025

adk-samples に学ぶデータ分析 LLM エージェント開発

Avatar for na0

na0

November 23, 2025
Tweet

More Decks by na0

Other Decks in Technology

Transcript

  1. from google.adk.agents import Agent from google.adk.tools import google_search agent =

    Agent( name="google_search_agent", model="gemini-live-2.5-flash-preview-native-audio-09-2025", tools=[google_search], instruction="You are a helpful assistant." ) adk-samples/python/agents/bidi-demo/
  2. data-science のツール • BigQuery クエリ実行 • AlloyDB クエリ実行 • BigQuery

    ML に関する知識検索 • 自然言語 to SQL • Python 実行
  3. 私たちの Agent v1 from google.adk.agents import Agent from google.adk.tools.bigquery import

    BigQueryToolset bigquery_toolset = BigQueryToolset(tool_filter=["execute_sql"]) root_agent = Agent( model="gemini-3-pro-preview", name="sample", instruction="You are a data science agent", tools=[bigquery_toolset], )
  4. 私たちの Agent v3 def rag_response(query: str) -> str: response =

    rag.retrieval_query( rag_resources=[rag.RagResource(rag_corpus=OUR_CORPUS)], text=query, ) return str(response) root_agent = Agent( tools=[bigquery_toolset, rag_response], // 変更のない引数略 )