Teradata DevTalk: Building Smarter AI Agents for Data Science Workflows

In this DevTalk, we’ll show how agents can support advanced data science workflows when they are guided not just on what to do, but on HOW to do best for a given platform. Agents and LLMs know how to accomplish overall business processes and analytic workflows with very little prompt engineering. The challenge is that they do not always choose the best way to execute those workflows for production settings. They default to writing client-heavy analytics code, perform unnecessary token/data movement, and use patterns that do not take advantage of the platform’s native strengths. In this session, we’ll use an open-source Teradata MCP project to show how agents can be guided to tap into native in-database analytics, AI and ML capabilities, and then deploy them into production-ready operational workflows. You’ll see how to move from simple prompts to more capable agent workflows that: identify the right analytic function for the task recognize when SQL alone is not enough use native in-database operators for prep, statistics, ML, text, and vector workflows chain these steps together into practical, end-to-end pipelines that can be easily deployed into production. Teradata includes built-in table operators for data preparation, statistics, machine learning, text analytics, and vector search. These functions run where the data lives, avoiding unnecessary data movement and taking advantage of Teradata’s parallel processing engine. Join Kevin Sturgeon, Solutions Engineer and Janeth Graziani, Developer Advocate, to see how skills, MCP Servers, and tools can help you build more capable AI agents for data science workflows. Bring your questions and follow along. Github Repo: https://github.com/ksturgeon-td/tdsql-mcp/blob/main/README.md tdsql-agent Github repo: https://github.com/ksturgeon-td/tdsql-agent Teradata Trial: https://www.teradata.com/getting-started/demos/clearscape-analytics Chapters: 00:00 Challenges with Agentic AI Systems Discussion of inefficiencies, hallucinations, and context issues in AI agents 01:49 Teradata SQL MCP Tool Overview Kevin Sturgeon introduces the open source project to guide agents for efficient analytics and native function usage. 07:25 Skills Framework & Model Context Protocol (MCP) Explanation of skills-based context injection, MCP server, and combining Teradata documentation for agent guidance. 14:35 Demo: Progressive Syntax Disclosure Live demonstration of the MCP server, guidelines, and native function discovery for agents. 18:56 Ad Hoc Analytics with Chat Tools Example of time series analysis, agentic workflow, and operational pipeline creation 25:23 Code Generation & Agentic Workflows Demonstration of code tools, error correction, and agent orchestration (LangChain, LangGragh) for unattended workflows. 31:35 Geospatial & Advanced Analytics Application of MCP tool for geospatial analysis and combining analytic techniques. 34:40 Project Access & Getting Started Instructions for viewers to access GitHub repo, Clearscape Analytics trial, and onboarding steps. 36:45 Q&A: Hallucinations, Prompt Accuracy, Agent Count Audience questions on prompt writing, hallucination reduction, agent setup, and accuracy verification. 43:30 Closing Remarks & Next Event Announcement Wrap-up, upcoming Teradata announcement