Graph over loop
Explicit nodes and edges reveal the permitted paths through reasoning.
MODULE 04 / 10
All modules ↑BUILD · TARGET L3 · 5–7 days
04
LangGraph agent workflow
Implement a read-only investigation graph with explicit state, termination, tool contracts and traceable decisions.
Introduce bounded probabilistic reasoning only where the workflow benefits from iterative evidence and hypothesis refinement.
Prerequisites
- Modules 01–03
- Python typing and testing
- Basic graph/state-machine concepts
01 / CONCEPT LESSONS
Build the mental model first.
Typed state
State should distinguish observations, hypotheses, evidence and recommendations.
Tool contracts
Validate inputs and return structured results; tools must not accept arbitrary commands.
Termination
Use step, time, cost and confidence limits with a safe escalation path.
Checkpointing
Resume from known state without repeating unsafe or expensive work.
02 / SYSTEM FLOW
Trace the complete path.
Normalize symptom→Collect signal→Form hypothesis→Select read-only tool→Evaluate evidence→Conclude or loop→Recommend / escalate
For every arrow, the engineer must name the data, identity, trust boundary, failure mode and evidence produced.
03 / GUIDED LAB
Build it in bounded steps.
AI and documentation are allowed. The learner owns every generated artifact and must preserve the stated safety boundaries.
- 01Define typed investigation state.
- 02Implement three read-only AWS tools.
- 03Add policy validation before every tool call.
- 04Set maximum steps, tokens and wall time.
- 05Checkpoint after tool results.
- 06Trace why the graph concluded or escalated.
04 / INCIDENT SIMULATION
The graph loops between two hypotheses and exhausts its token budget.
Mentor injects
- Confidence never changes
- Tool returns the same evidence
- Termination checks run only after model calls
Required evidence
- Loop trace
- Budget consumption timeline
- New convergence and escalation rule
Investigation loop
Symptom → impact → hypothesis → evidence → test → root cause → containment → correction → verification.
05 / DESIGN CHALLENGE
Start with a blank page.
Decide which investigation steps remain deterministic and which become graph nodes. Defend every probabilistic edge.
06 / VERIFY & SUBMIT
Verification checklist
Submission package
- Graph diagram
- State schema
- Tool contracts
- Loop incident report
- Trace excerpt with explanation
Review questions
- Why is this edge probabilistic?
- What stops oscillation?
- Can a tool result contain hostile text?
- What is persisted?
- When does the graph escalate?
07 / MENTOR-ONLY GUIDANCE
Randomly remove one stop condition during review and ask the engineer to predict the exact operational effect.
SCORING1 · cannot explain2 · explains with gaps3 · implements and verifies4 · designs and handles trade-offs
Record the achieved L-level only from observed evidence. Course completion and certification do not automatically change the skill matrix.