vapi_skill

Overview

This skill provides mapping guidance to convert Vapi voice event fields into dashboard-ready templates. It standardizes common voice event vocabulary (call, transcript, duration, cost, status) and supplies heuristics to map inconsistent field names to canonical dashboard properties. The output focuses on a voice-analytics template for quick integration into reporting pipelines.

How this skill works

The skill inspects incoming Vapi voice event payloads and applies a small set of heuristics to resolve field name variations into canonical properties: duration, status, and cost. It outputs a normalized object matching the voice-analytics template, preserving raw fields for auditability. Use the mapping rules to transform events before storing, aggregating, or feeding visualization layers.

When to use it

• Ingesting heterogeneous Vapi voice event streams with inconsistent field names.
• Preparing data for a voice-analytics dashboard that expects canonical fields.
• Normalizing events before cost, duration, or status-based aggregation.
• Building ETL pipelines where downstream systems require stable property names.
• Validating or enriching call records prior to long-term storage.

Best practices

• Apply the mapping heuristics early in the pipeline to avoid branching logic downstream.
• Keep original event fields alongside canonical fields for traceability and debugging.
• Treat mappings as configurable rules so you can extend or override heuristics for edge cases.
• Validate numeric conversions (e.g., duration, cost) and handle missing or malformed values explicitly.
• Log mapping decisions at a coarse level to monitor changes in upstream schemas.

Example use cases

• Normalize events where call duration appears as call_duration_seconds, duration, or call_length into a single duration property.
• Map different status fields (status, call_status, outcome) to a canonical status used by the dashboard.
• Aggregate daily call cost by mapping cost_usd, cost, or price into cost and summing per call.
• Feed normalized voice-analytics objects into a real-time metrics pipeline for live dashboards.
• Create alerts based on normalized status transitions (e.g., FAILED, DROPPED, COMPLETED).