Building and Evaluating Data Agents
by DeepLearning.AI
Build a planner-driven multi-agent data agent in LangGraph, then actually measure whether it works.
Overview
Building and Evaluating Data Agents is a DeepLearning.AI short course built in partnership with Snowflake and taught by Anupam Datta, AI Research Lead at Snowflake, and Josh Reini, a Snowflake developer advocate. It runs 1 hour 59 minutes across eight video lessons with five code examples and a graded quiz, and is pitched at an intermediate level. The build target is a data agent that answers questions over both structured and unstructured sources: you assemble a LangGraph multi-agent workflow with a planner that decides which sub-agents to call and in what order, a plan executor, Cortex sub-agents that retrieve from Snowflake, web search for anything outside the warehouse, and further sub-agents that summarise or visualise the result. The second half is evaluation, and it is the real subject of the course. You add tracing to the workflow, then score it on two axes: output quality, using LLM-as-a-judge metrics for context relevance, groundedness and answer relevance; and trustworthiness, using alignment metrics across the agent's goal, plan and actions — plan quality, adherence, consistency and efficiency. That goal-plan-action framing is the same one Snowflake documents for its own Cortex Agent evaluations, which likewise compute metrics with an LLM-as-a-judge technique. The closing lessons shut the loop by revising the agent's prompt and adding inline, real-time evaluations. Access is free during the DeepLearning.AI platform beta; the graded assignment and certificate require a Pro membership.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Course
- Cost
- Freemium
- Duration
- 1 hour 59 minutes, self-paced (8 lessons + quiz)
- Provider
- DeepLearning.AI
- Hands-on
- Yes — code/exercises
- Certificate
- Available
What You’ll Learn
- ✓Design a planner that chooses which sub-agents to call and in what order
- ✓Assemble a multi-agent workflow in LangGraph with a separate plan executor
- ✓Add Cortex sub-agents to retrieve structured and unstructured Snowflake data
- ✓Extend a data agent with web search, summarisation and visualisation sub-agents
- ✓Instrument the whole workflow with tracing so failures are attributable
- ✓Score output with LLM-as-a-judge context relevance, groundedness and answer relevance
- ✓Compute goal-plan-action alignment metrics for plan quality, adherence, consistency and efficiency
- ✓Close the loop by revising prompts and adding inline real-time evaluations
Highlights
- •Evaluation is the curriculum, not an afterthought — three of the eight lessons are measurement
- •Teaches goal-plan-action alignment, which catches an agent that reached the right answer the wrong way
- •Taught by Snowflake's AI research lead and a developer advocate, using the framing Snowflake ships in its own product docs
- •Concrete stack rather than pseudocode: LangGraph for orchestration, Snowflake Cortex for retrieval, LLM-as-a-judge for scoring
- •Under two hours with five runnable code examples, so it fits an afternoon
- •Vendor-shaped, and worth knowing going in: the retrieval path is Snowflake Cortex, so the orchestration and evaluation patterns transfer more readily than the code
Who It’s For
Best For
- ✓Engineers building agents over enterprise databases and documents
- ✓Practitioners who can build an agent but cannot yet prove it is reliable
- ✓Data teams already on Snowflake who want an agent layer over Cortex
- ✓Anyone learning LangGraph who wants a non-toy multi-agent example
Prerequisites
- •Basic Python and object-oriented programming
- •Familiarity with LLM prompting and the idea of tool calling
- •Helpful but not required: exposure to SQL or a data warehouse
FAQ
What is Building and Evaluating Data Agents?
A two-hour DeepLearning.AI short course built with Snowflake that has you assemble a LangGraph data agent — planner, executor and specialised sub-agents over databases, files and web search — and then instrument and score it. The evaluation half is the point: tracing, LLM-as-a-judge output metrics, and goal/plan/action alignment metrics.
Is Building and Evaluating Data Agents free?
Building and Evaluating Data Agents offers free content, with paid options for certificates or premium features.
What level is Building and Evaluating Data Agents for?
Building and Evaluating Data Agents is aimed at a intermediate audience. Recommended background: Basic Python and object-oriented programming, Familiarity with LLM prompting and the idea of tool calling, Helpful but not required: exposure to SQL or a data warehouse.
How long does Building and Evaluating Data Agents take?
Expect roughly 1 hour 59 minutes, self-paced (8 lessons + quiz). Most learners work through it at their own pace.
What will I learn from Building and Evaluating Data Agents?
You'll learn: Design a planner that chooses which sub-agents to call and in what order; Assemble a multi-agent workflow in LangGraph with a separate plan executor; Add Cortex sub-agents to retrieve structured and unstructured Snowflake data; Extend a data agent with web search, summarisation and visualisation sub-agents; Instrument the whole workflow with tracing so failures are attributable; Score output with LLM-as-a-judge context relevance, groundedness and answer relevance; Compute goal-plan-action alignment metrics for plan quality, adherence, consistency and efficiency; Close the loop by revising prompts and adding inline real-time evaluations.
Topics
Sources
This page was written from 3 sources, 2 on domains other than deeplearning.ai.