Phone Agent Training Pipeline

Synthetic conversation corpus for training a plumbing company phone agent. 170 multi-turn conversations covering 8 problem types, 7 caller personalities, and 8 conversation paths with full phase annotations.

170 Conversations
2,319 Total Turns
11 Phases
128 21 21
Train / Val / Test

Overview

This dataset trains a phone receptionist agent for Smith Plumbing. Each conversation is annotated with per-turn phase labels, enabling the model to learn both natural dialogue flow and the structured phases of a service booking call.

Problem Types

8 plumbing scenarios: leak repair, clogged drain, toilet repair, water heater, faucet install, garbage disposal, sewer line, and emergency calls.

Caller Personalities

7 personality archetypes: default, rushed, chatty, angry, confused, elderly, and non-native English speakers. Each shapes tone and pacing.

Conversation Paths

8 resolution paths: happy path booking, no available slots, emergency escalation, DIY recommendation, question only, cancellation, caller hangup, and changes mind.

Phase Annotations

Every turn is labeled with its conversation phase, enabling phase-aware training and evaluation of the agent's ability to navigate structured service calls.

Conversation Phases

Each turn in every conversation is annotated with one of 11 phases. Click a phase to filter the explorer below to conversations containing that phase.

Conversation Explorer

Browse and search all 170 training conversations. Filter by problem type, personality, path, or search transcript text. Click a conversation to read the full transcript.

Dataset Statistics

Distribution of conversations across problem types, personalities, paths, and phase coverage heatmap.

Problem Types

Personalities

Conversation Paths

Phase Coverage