Path 01 / Beginner
AI Foundations
Understand how modern AI works and build useful AI-powered applications.
- Language models
- Prompt design
- Tokens & context
- Structured outputs
- Model limitations
- Verification & evaluation
AI Lab / Learn. Build. Experiment.
Don’t just learn AI. Build it.
Explore AI agents, automate workflows, and build practical AI systems through hands-on projects and interactive experiments.
Interactive simulation / 24 × 18
ready
Node inspector / Pending
User request
- Input
- What is 24 × 18?
- Action / decision
- Receive a supported sample calculation.
- Output
- Expected: Calculate 24 × 18.
- Possible failure
- Unsupported or empty request.
Execution trace
- Awaiting execution.
Predefined tool selection and local arithmetic. No live model or external action.
01 / Learn
Choose your path.
Build your understanding step by step, from AI fundamentals to engineering complete AI systems.
Path 02 / Beginner to intermediate
Workflow Automation
Connect applications, process information, and design reliable automated workflows.
- Triggers & actions
- Conditions & branching
- APIs & webhooks
- Data transformation
- Error handling
- Human approval
Path 03 / Intermediate
AI Agent Engineering
Explore tool selection, observations, state, and bounded execution.
- Tool calling
- Planning
- Memory & state
- Execution limits
- Guardrails
- Evaluation & debugging
Path 04 / Intermediate to advanced
AI Systems Engineering
Connect models, retrieval, APIs, and application logic into complete solutions.
- Python & FastAPI
- Database fundamentals
- Embeddings
- Semantic search & RAG
- Architecture
- Testing & deployment
Existing curriculum & saved learning paths
- PATH 01Published
AI Agent Fundamentals
Understand how AI agents combine models, instructions, tools, and controlled execution to complete tasks.
- Distinguish models, agents, and automation
- Understand agent components and tool calling
- Identify use cases, permissions, and evaluation needs
Beginner · 6 published lessons
Explore path - PATH 02Coming soon
Build AI Agents with Python
Build working agents by connecting language models to Python applications and approved tools.
- Set up Python and connect a model
- Implement tool calling and conversation state
- Handle errors and evaluate behavior
Beginner to intermediate · 0 published lessons
Explore path - PATH 03Coming soon
APIs and Business Integrations
Connect agents to APIs, databases, and business workflows using controlled integrations.
- Understand APIs and structured outputs
- Expose application functionality with FastAPI
- Use isolated training data, authentication, and validation
Intermediate · 0 published lessons
Explore path - PATH 04Coming soon
Local AI and Private Agents
Explore locally running models, document retrieval, and private AI application architectures.
- Compare cloud, local, and hybrid deployments
- Connect Python to Ollama
- Understand embeddings, vector search, and retrieval evaluation
Intermediate · 0 published lessons
Explore path - PATH 05Coming soon
Production AI Engineering
Learn the engineering practices needed to test, monitor, secure, and maintain AI systems.
- Design execution boundaries and handle untrusted inputs
- Introduce evaluation, observability, and human approval
- Measure costs and latency; plan maintenance
Advanced · 0 published lessons
Explore path
02 / Experiment
See how an AI agent thinks through a task.
Explore the workflow, inspect the tools, change the instructions, and see how the system responds.
Finance agent / sample environmentRead-only tools
Text edits are checked for a supported read-only invoice summary. The controls below determine the actual calculation.
Inspect all five sample invoices
| Invoice | Days overdue | USD | Disputed |
|---|---|---|---|
| INV-101 | 45 | $1,200.00 | No |
| INV-102 | 30 | $800.00 | No |
| INV-103 | 65 | $2,400.00 | Yes |
| INV-104 | 31 | $750.00 | No |
| INV-105 | 12 | $500.00 | No |
ready
Node inspector / Pending
Interpret task
- Input
- Find invoices that are overdue by more than 30 days and prepare a summary for the finance team.
- Action / decision
- Validate supported read-only intent; configurable controls determine the rule.
- Output
- Expected: Review the available invoice records, identify invoices overdue by more than 30 days, calculate the outstanding total, and prepare a concise summary. Do not send messages or modify records.
- Possible failure
- Empty, unsupported, or write-action request.
Execution trace
- Awaiting execution.
This interactive environment uses sample data and simulated tool execution. It does not access real financial records or perform external actions.
03 / Build
Build something real. One step at a time.
Follow guided projects, understand the architecture, and extend each system with your own ideas.
6 projects
Project 01 / AI Agents
Build Your First AI Agent
Select an approved tool, inspect its result, and return a verified answer.
- Instructions
- Tool selection
- Execution traces
Project 02 / AI Applications
Intelligent Document Processor
Extract sample invoice fields, correct an exception, and rerun validation.
- Structured extraction
- Validation
- Exceptions
Project 03 / Data and Retrieval
Build a Knowledge Assistant
Retrieve supplied evidence and distinguish grounded answers from unsupported questions.
- Retrieval
- Grounding
- Source verification
Project 04 / Automation
Email Triage Automation
Classify fictional inquiries and hold consequential routing for approval.
- Classification
- Branching
- Human approval
Project 05 / Data and Retrieval
AI Research Assistant
Produce a briefing from supplied passages while identifying gaps in the evidence.
- Decomposition
- Evidence checking
- Structured reports
Project 06 / Automation
Workflow Debugger
Trace a failed workflow, identify the cause, and correct its configuration.
- Logs
- Failure handling
- Retries
04 / Automation studio
Understand the workflow. Improve the system.
Compare rule-based automation, AI-assisted workflows, and tool-using agents using the same sample inquiry.
Predefined keywords select a route. Ambiguous requests need escalation; a simpler automation is often sufficient.
Sample inquiry: “Please help me understand your automation offering.”
ready
Node inspector / Pending
Trigger
- Input
- Sample inquiry: Please help me understand your automation offering.
- Action / decision
- Simulate trigger using the supplied inquiry.
- Output
- Expected: Sample step completed; no external action.
- Possible failure
- Missing destination; stop before subsequent actions.
Execution trace
- Awaiting execution.
Compare engineering trade-offs
| Characteristic | Rule-based automation | AI-assisted automation | Tool-using AI agent |
|---|---|---|---|
| Predictability | High for defined rules | Rules constrain model proposals | Bounded, but decisions vary |
| Flexibility | Known patterns only | Handles varied phrasing | Adapts using observations |
| Complexity | Low | Moderate | Higher |
| Model inference | Not needed | Classification proposal | Tool-selection decisions |
| Verification | Rule and route checks | Schema and policy checks | Every tool result and final output |
| Human oversight | Escalate exceptions | Approve consequential routing | Approve consequential actions |
All modes are predefined simulations. No model inference, real classification, or routing is performed.
05 / Challenge board
Can you solve the problem?
Practice one engineering decision at a time. Challenge answers stay in this session.
Mission 01 / Practice
Choose the Right Tool
A user asks for 24 × 18. Which permitted tool should handle it?
Extension challenge
What should happen if the calculator rejects an argument?
Mission 02 / Practice
Fix the Broken Workflow
The trace stops at Route: destination missing. What should change?
Extension challenge
Which failures are safe to retry?
Mission 03 / Practice
Find the Missing Evidence
The source only says “support replies within two business days.” Which answer is unsupported?
Extension challenge
Rewrite the unsupported answer with an explicit uncertainty.
Mission 04 / Practice
Add Human Approval
A workflow proposes sending a customer message. Where should approval happen?
Extension challenge
What other actions require a review gate?
Mission 05 / Practice
Validate the Output
Required schema: invoice_id string, amount positive number. Which output passes?
Extension challenge
Add an allowed-currency check.
Mission 06 / Practice
Handle Tool Failure
A read tool times out. What is the safest bounded response?
Extension challenge
Set a maximum retry count and escalation policy.
06 / System gallery
Explore what you can build.
Inspect educational system concepts, their boundaries, and related sample demonstrations.
Data and Retrieval / concept
Research Assistant
AI Applications / concept
Document Intelligence System
Data and Retrieval / concept
Knowledge Assistant
AI Agents / concept
Customer Operations Agent
AI Applications / concept
AI Data Analyst
Workflow Automation / concept
Workflow Debugger
Advanced AI Systems / concept
Multi-Agent Research Workflow
Workflow Automation / concept
Human-in-the-Loop Automation
07 / How the lab works
Learn. Build. Test. Improve.
- Learn the concept
- Build a project
- Experiment with rules
- Test failure cases
- Improve the system
Find your next step.
7 published results
lesson · Beginner
What Is an AI Agent? →
Understand what an AI agent is, what it does, and how it differs from a traditional chatbot.
lesson · Beginner
AI Models vs. AI Agents vs. Automation →
Tell apart AI models, AI agents, and traditional automation — and learn when each is the right tool.
lesson · Beginner
How an AI Agent Works →
Follow the execution cycle of an AI agent, step by step, from request to checked result.
lesson · Beginner
Tools, Instructions, and Context →
Learn how instructions, context, tools, and permissions shape everything an agent does.
lesson · Beginner to intermediate
Understanding Agent Architecture →
Meet the components of a simple agent application and see how they communicate.
lesson · Beginner to intermediate
When Should You Use an Agent? →
Learn to evaluate use cases, weigh the trade-offs, and decide whether an agent is worth building.
project · Beginner
Build Your First AI Agent →
Build a small Python agent that calls one approved read-only tool and returns a grounded task status.
Your next step
Your next AI system starts here.
Choose a project, explore a workflow, or start learning how AI agents and automation work.