AI catalog search

Optional track — unlike Use Case 12: Copilot, categorizer and lookup live in the same training bot. A separate Copilot bot is not required.

Required permissions and access

Module / setting	Why you need it
Scenario Builder	FAQ + categorizer + CD lookup in one bot
Fast Line Pro	Categorizer or intent for the catalog
Custom Data	`products` model + seed
Settings → Operator Panel	Handoff to an operator (optional)
Runs / Chat preview	5 control queries

Business context

The customer asks for a product or character from the training catalog. The bot replies from Custom Data and, if needed, hands the dialog to an operator. A typical pattern for e-commerce / reference bots without a Copilot architecture.

Expected result

Model products + seed products-training.csv
Categorizer or filter in CD after action_fastline_pro
Handoff edge on “not found” or at the customer’s request
Not a separate Copilot bot — everything in the training bot

Architecture

MessageKeyboard "Catalog" → WaitForInput (query)
   → action_fastline_pro (categorizer / intent)
   → Router (valid filters)
   → custom_modules__get / search
      ├─ found     → MessageKeyboard (result)
      ├─ not_found → clarification or handoff
      └─ error     → message + fallback

Optionally: "Operator" → connect (Use Case 3)

Step-by-step implementation

Step 0. Custom Data — products

Create the products model (alias products).
Fields: name, category, faction, price.
Import products-training.csv.

→ Custom Data set

Step 1. Fast Line Pro — catalog categorizer

Parameter	Value
Type	Chatbot
Prompt	Catalog section in categorizer-training.txt
Test	“spellcaster from north” → JSON filters

Step 2. Entry and categorizer in the scenario

Node	Parameters
MessageKeyboard	“What are you looking for in the catalog?” → WaitForInput
WaitForInput	outputVariable: `catalog_query`
action_fastline_pro	`user_input`: `{{catalog_query}}`, `save_response`: `catalog_filters`

Edges: main → lookup; fallback → “Please rephrase your query.”

Step 3. Lookup in Custom Data

node_params (example)	Value
`module_name`	`products`
filter by	`category`, `faction` from categorizer

Edges:

Edge	Message (example)
found	“{{name}} — {{price}} gold”
not_found	“Not found in the catalog. Try differently or tap “Operator”.”
error	“Search error.”

Step 4. Handoff to an operator

On not_found or the “Operator” button:

MessageKeyboard → operator_panel__connect_to_operator_with_msg
   (subject_alias from Use Case 3)

Edges: success / limit / error — required.

→ Use Case 3

Step 5. Five control queries (self-check)

#	Query	Expectation
1	“Geralt”	found — warrior, north
2	“dragon from Mars”	not_found
3	“something cheap”	found or clarification (not silence)
4	“Operator” / handoff	connect edges
5	Unknown intent	categorizer fallback

Each pass — a separate Run or reset state; record the edge in Runs.

Troubleshooting

Symptom	What to check
Always not_found	Filters vs seed CSV; case of category/faction
AI “invents” a product	Lookup must come from CD, not only from LLM text
Endless clarifications	Retry limit → handoff
Handoff without limit edge	Edge-driven flow

Self-check

5 control queries passed; results in the result template
Found / not found / handoff / fallback — all with a message to the customer
module_name = CD alias
You understand the difference from Use Case 12 (Copilot vs inline)

Independent task

Add MessageKeyboard after found: “Search again” → loop to WaitForInput; “To menu” → capstone main menu.