Agent Builder Path
Six lessons covering tool definition · actors + HITL · RAG · intent gating · debugging — the full cycle of designing agent workflows that safely unpack natural-language input into tools and decisions.
0/6 complete
About this Path
This Path walks the cycle of designing agent workflows that unpack natural-language input into tool calls and decisions in D.Hub. Tool definition · actors · RAG · intent gating · debugging — six lessons, about 55 minutes total.
The audience is the builder during AI automation adoption. If an analyst is in the seat that uses the AI Assistant, the learner of this Path is in the seat that directly designs an agent for a specific organizational workflow.
Prerequisites
- A D.Hub portal account with engineer or analyst access (Editor or higher)
- One of the Essentials Path or the Analyst / Engineer Paths finished — collection · dataset · pipeline basics need to feel familiar before the patterns here land
- Scenario zips for the refund_approval, docai, and map_control workshops on hand — when you have them, each lesson's example works immediately
What you'll be able to do
- Explain the three axes of an agent — Tool · Actor · Intent classification.
- Define a new tool with an input/output schema contract.
- Separate actors into automated vs human-review, and design a HITL workflow.
- Pull knowledge resources via RAG (embedding search) and require citation grounding in responses.
- Apply default deny safety through intent classification + allow-list gating.
- Operate the agent with call logs (e.g.
om_event_history) and rejection-rate dashboards as monitoring signals.
What comes after this Path
The 6 lessons of this Path are the first full cycle of the agent builder. To go deeper:
- Three workshops — refund_approval · docai · map_control — walk the full flow end-to-end so you can see how each pattern fits within a real scenario.
- Admin Path — When an agent rolls out organization-wide, permission · policy · audit surfaces stack on top.
- Ontology Modeler Path — Modeling for when the agent's knowledge grounding extends to a graph base.
Check off each lesson as you finish it — progress is recorded automatically.
Lessons
- 01Agent overview — the three axes of Tool · Actor · IntentAlign D.Hub agents on their three axes — Tool · Actor · Intent classification — and the 5-step one-cycle flow.8 min
- 02Tool definition — the input/output schema contractThe flow for defining one tool — name · description · input parameters · output schema · the *when does the LLM call this tool* decision signal.9 min
- 03Actor + HITL — automated vs human-review actorsThe pattern of separating actors into automated and human-review forms, designing a HITL workflow of *recommend → human decide → automated execution*.10 min
- 04RAG + knowledge grounding — embedding search and citationPull knowledge resources via RAG (retrieval-augmented generation) tools and require the LLM response to carry *citation grounding*.9 min
- 05Intent classification + allow-list gating — the default-deny safety patternFunnel natural-language input through intent classification → allow-list gating to apply the *default deny* safety pattern. Generalizing the map_control workshop pattern.9 min
- 06Debugging + monitoring — call logs + rejection-rate dashboardA debugging flow built on agent call logs (e.g. om_event_history) and a monitoring dashboard pattern for rejection rates · HITL agreement rates.10 min