Data Agents & Intelligent Analytics Systems
Design governed Data Agents that understand analytical intent, access data/tools safely, execute and verify analysis, explain evidence, and leave auditable traces: start by deciding when an agent is justified, translate requests into an analysis plan before executing, give the agent narrow allowlisted tools with read/write separation, ground metrics in a trusted semantic layer, build a verifiable read-only prototype in a Guided Lab, then learn to propagate authorization and minimize data exposure, verify answers with evidence before presenting them, evaluate with golden questions and adversarial tests, add observability and human-in-the-loop approval, and deliver a governed Data Agent architecture, prototype, and evaluation pack in an Independent Lab.
The course storyline: from a chatbot with data access to a governed Data Agent
The whole course follows one learning loop: Intent, Plan, Tool/Data Access, Execute, Verify, Explain, Approve, Observe. AI may plan and execute within governed bounds; a person stays responsible for deciding whether an agent is justified, approving consequential actions, and owning the result.
- 1. Intent
What is this request actually asking for, and does it genuinely need an agent?
- 2. Plan
Have the entities, metrics, filters, time range, and evidence requirements been resolved before execution?
- 3. Tool/Data Access
Does the agent use only narrow, allowlisted, read/write-separated tools?
- 4. Execute
Was the query/code actually run, with the dataset version recorded?
- 5. Verify
Does the answer pass a verification harness (query scope, row counts, reconciliation)?
- 6. Explain
Can the answer cite its evidence, dataset version, and stated uncertainty?
- 7. Approve
Is there a human approval point before a high-impact write/publish action?
- 8. Observe
Are intent, plan, tool calls, versions, and results being logged for audit?
Your next step
DS 302 · Module 1: Data Agent Architecture: Intent, Tools, Data, and Semantics
Curriculum
- Module 1Data Agent Architecture: Intent, Tools, Data, and SemanticsStart by deciding when an agent is justified at all, translate a request into an analysis plan before executing anything, give the agent narrow allowlisted tools, ground its metrics in a trusted semantic layer, then build a verifiable read-only prototype in a Guided Lab.
- Module 2Reliable Intelligent Analytics: Verification, Permissions, Evaluation, and GovernanceLearn to propagate authorization and minimize data exposure, require evidence before an answer is trusted, build a repeatable evaluation suite against adversarial and ambiguous requests, add observability and human-in-the-loop approval, then deliver a governed Data Agent architecture in an Independent Lab.
Progress is stored in this browser and does not identify you. It is uploaded only if you join progress sync below.
Sync my progressJoin a session to keep progress across devices and share it with your instructor
Practice and assessment
- Practice A: Agent-vs-Workflow DecisionDecide which sample requests genuinely need a dynamically planning agent versus a fixed dashboard or workflow.Not done yet
- Practice B: Analysis Plan CritiqueCritique an underspecified analysis plan and rewrite it so every required element is explicit.Not done yet
- Practice C: Tool Contract DesignDesign narrow, allowlisted tool contracts with explicit schemas and read/write classification.Not done yet
- Practice D: Semantic Metric ResolutionResolve two conflicting ad hoc metric definitions into one semantic-layer entry.Not done yet
- Practice E: Permission Boundary TestTest whether a sample agent tool correctly enforces row/column security for a role.Not done yet
- Practice F: Golden Question EvaluationBuild a small golden-question set and check a sample agent response for regressions.Not done yet
- Practice G: Prompt Injection DefenseIdentify an instruction hidden in retrieved data and verify it is treated as untrusted content.Not done yet
- Practice H: Verification Harness DrillRun a sample answer through a verification harness checklist before it is presented.Not done yet
- DS 302 pre-test and post-testPre- and post-tests check how your judgment changes on when to use an agent, planning, tool design, read/write separation, semantic layers, security, verification, prompt injection, evaluation, observability, failure recovery, and human-in-the-loop approval.Not done yet
- Data Agent Architecture Canvas + Governed Prototype + Evaluation PackTurn what you learned into your own Data Agent Architecture Canvas + Governed Prototype + Evaluation Pack, stating the problem, data, method, verification, and governance judgment.Not done yet
About this course
Outcomes
- Define when an agent is justified; distinguish natural-language interface from autonomous analytical action
- Translate request into entities, metrics, filters, time range, plan, evidence requirements, approval needs
- Use allowlisted tools with narrow schemas; separate read from write/action tools
- Ground metrics, dimensions, definitions, join paths, units, and policy in a trusted semantic layer
- Build a read-only prototype that plans, queries, executes Python if needed, cites dataset versions, and verifies totals
- Complete the final artifact: Data Agent Architecture Canvas + Governed Prototype + Evaluation Pack
Prerequisites
- DS 301 and AI 104 recommended; AI 201 also recommended
- DS 101–301 level Python/pandas skills expected
Who it is for
- NTPU students, faculty and administrative staff who work with structured data
- Learners who completed DS 301 and want to design governed, verifiable Data Agents
- Analysts who need stronger verification, permission governance, and responsible AI/data practice
Certificate
Earns a course completion certificate (L1); a capability certificate (L2) is also available through additional artifact review.
Requirements to earn it
- Complete all 2 course modules
- Pass the post-test assessment at 80% or higher
- Submit a project artifact and pass review
Course feedback
It takes 2 to 3 minutes and is used to improve the course.