Arbitra

From Call to Billable Evidence

Arbitra ingests calls from your call platform, transcribes and evaluates using your providers, then applies deterministic scoring policies.

Architecture

Evaluation Pipeline

Ingestion

Call platform + recordings

Transcribe

AssemblyAI (default)
Other providers

Evaluate

Your Rules
OpenAI
Anthropic
AWS Bedrock (by request)

Reports

Scores, evidence, and exports

The Four-Stage Pipeline

STEP 01

Call Ingestion

Calls sync from your call platform with recordings and metadata.

  • Connect call platforms and recording tools (GoHighLevel, OpenPhone, RingCentral, and more)
  • Pulls call metadata and recording links automatically
  • Tracks call status through transcription and evaluation
  • Designed for multi-client agency workflows
STEP 02

Transcription

Transcription with diarization and timestamps.

  • Speaker diarization (who said what, when)
  • Word-level timestamps for review and evidence
  • Bring your own transcription provider (AssemblyAI supported today)
  • Dual-channel / wideband audio improves diarization when available
  • Mono 8kHz call audio has inherent accuracy limits
  • Transcription is independent from scoring logic
STEP 03

Criteria-Based Evaluation

Provider-agnostic evaluation of your criteria and rules.

  • Define criteria with weights, types, and rules
  • Each criterion is evaluated independently
  • Provider keys live in Settings and override env defaults
  • OpenAI and Anthropic supported today; other providers available by request for private deployments (e.g., AWS Bedrock)
  • Cross-provider fallback on retryable failures
STEP 04

Deterministic Scoring + Reporting

Business rules applied consistently with client-ready exports.

  • Deterministic scoring policy for billable, tier, and outcome
  • Review reasons and manual overrides for edge cases
  • PDF and CSV exports for client billing workflows
  • Evaluation history and scoring changes retained

Operational Guardrails

30s
Short-Call Rule
Auto-mark short calls as non-billable
Strict
Contract V2
Missing required fields mark calls incomplete
On
Deterministic Policy
Same inputs always produce the same score
Built in
Manual Review
Override billable decisions when needed

Provider Settings That You Control

Bring your own keys. Choose models per provider. Settings in the app override env defaults.

Evaluation Providers

Provider-agnostic evaluation layer. OpenAI and Anthropic are supported today; privacy-first deployments can use providers like AWS Bedrock by request.

  • OpenAI (GPT models)
  • Anthropic (Claude models)
  • AWS Bedrock (by request)
  • Cross-provider fallback on retryable errors

Transcription Provider

Transcription provider is configurable. AssemblyAI is supported today, and private deployments can use other providers by request. Accuracy depends on recording quality and format.

  • AssemblyAI transcription (default)
  • Speaker diarization
  • Dual-channel/wideband audio recommended
  • Bring your own provider key/credentials

See the pipeline in action

Reserve a Private Deployment Sprint to see a call go from audio to scored evidence with your criteria.

Request Private Pilot