AI Agents and Jupyter Notebooks for Security Data Analysis
summary
Claim: letting an AI agent run Python inside a live Jupyter notebook to load query results as a DataFrame and do post-query filtering, grouping and visualization reduces back-and-forth with the LLM and makes analysis repeatable. Implemented an MCP server plus two Python packages (Jupyter toolkit and an agent data toolkit with posgress tools) that run locally or remote, connect to Postgres and Jupyter, execute code cells and produce visuals (heat map) over a 4,000‑row Defender dataset, surface a brute‑force IP, and expose reasoning as markdown audit trail while using oppus 4.6 in demos.
AI-written summary · about
chapters (11)
- 0:00AI agents and Jupyter notebooks
- 3:00LLM data analysis limits
- 6:13Jupyter notebooks as AI bridge
- 11:36Kernel and notebook integration
- 15:47Jupyter demos local and remote
- 22:10Agents connecting to PostgreSQL
- 31:02Agent database toolkit
- 39:33Security data analysis case
- 49:08Agent workflows with AI skills
- 59:24Live structured data analysis
- 1:01:34Closing thoughts and future
related talks
28m"Cyber Security Investigations with Jupyter Notebooks"
39mJupyter AI — Bringing Generative AI to Jupyter
01h 04mBuilding AI Applications the Pydantic Way
38mSponsored Session: Jupyter AI v3: Human + AI Collaboration
42mSimplifying Data Analysis with GitHub Codespaces, Jupyter Notebooks & Open AI
23mFine Tune Your AI Toolbox with GitHub Copilot, Azure AI Foundry...
45mAgents, Tools, and MCP, oh my! Next-level AI concepts for developers
54mThe secret weapon for developers, Notebooks.