Back to HomePlatform Architecture
Home/Product Architecture

Platform

VOCAL Product Architecture

VOCAL uses a layered architecture to ingest conversation data, generate transcripts and structured AI findings, calculate call analytics KPIs, and deliver action-ready outputs to dashboards, APIs, and downstream systems.

Core Data Flow

  1. 1. Stage 1 - Call Source and Ingestion

    Capture telephony recordings, uploaded media, or connected source-system calls; normalize metadata and start processing jobs.

  2. 2. Stage 2 - Transcription and Segmentation

    Generate timestamped transcripts, speaker-aware segments, and confidence-aware normalization artifacts.

  3. 3. Stage 3 - AI Analysis and Signal Detection

    Detect sentiment, intent, objections, urgency, opportunity and risk cues, plus QA/compliance language patterns.

  4. 4. Stage 4 - KPI Mapping and Scoring

    Map detected signals into rubric scores, KPI measures, and role-specific scorecard components.

  5. 5. Stage 5 - Structured Persistence

    Persist normalized records for calls, transcripts, findings, entities, metrics, and processing state.

  6. 6. Stage 6 - Delivery Layer

    Deliver results to dashboard views, call detail workflows, API consumers, export feeds, and downstream systems.

Architecture Walkthrough

Each stage has explicit responsibilities, structured handoffs, and downstream dependencies. This is how VOCAL keeps dashboard, API, and export outputs aligned to the same processing model.

Stage 1 - Call Source and Ingestion

Create a reliable input contract so every downstream stage has consistent call and source context.

Primary input
Audio/recording reference plus source identifiers and call metadata.
Primary output
Validated call record, normalized metadata, and orchestration-ready processing state.
Why it matters
Ingestion quality controls attribution accuracy, routing, and score reliability later in the pipeline.
Connects next to
Transcription and segmentation.

Stage 2 - Transcription and Segmentation

Convert raw audio into searchable, analyzable conversation structure.

Primary input
Normalized call media and ingestion context.
Primary output
Transcript text, utterance boundaries, speaker turns, and transcript confidence signals.
Why it matters
Every signal, metric, and coaching output depends on transcript and speaker quality.
Connects next to
AI analysis and signal detection.

Stage 3 - AI Analysis and Signal Detection

Interpret conversation behavior into explainable, structured findings.

Primary input
Transcript segments, speaker context, and call-level metadata.
Primary output
Findings/entities with evidence context suitable for KPI mapping and review workflows.
Why it matters
Transforms conversation text into decision-grade operational intelligence.
Connects next to
KPI mapping and scoring.

Stage 4 - KPI Mapping and Scoring

Standardize interpretation so teams evaluate calls consistently.

Primary input
Structured findings/entities plus KPI and rubric logic.
Primary output
Call-level and aggregate metric outputs with score-ready structures.
Why it matters
Enables coaching, QA, and operational monitoring from common definitions.
Connects next to
Structured persistence.

Stage 5 - Structured Persistence

Create an analytics-ready source of truth for retrieval, export, and trend analysis.

Primary input
Stage outputs from ingestion through scoring.
Primary output
Searchable and exportable call-intelligence records with governance context.
Why it matters
Prevents one-off insights by making outputs auditable and reusable across teams.
Connects next to
Delivery layer (dashboard, API, exports, downstream systems).

Stage 6 - Delivery Layer

Operationalize intelligence where business decisions are made.

Primary input
Structured persisted call-intelligence records.
Primary output
User-facing insights, machine-readable API payloads, and integration-ready exports.
Why it matters
Connects analysis quality directly to sales, QA, compliance, and operations outcomes.
Connects next to
Continuous monitoring, governance, and business workflows.

System Layers

Ingestion Layer

Captures audio, call metadata, and source-system context for downstream processing.

Transcription Layer

Transforms audio into timestamped text with speaker-aware segmentation artifacts.

Analysis Layer

Detects semantic, emotional, behavioral, and policy-relevant conversation signals.

Scoring Layer

Calculates KPIs, quality metrics, and role-specific evaluation outputs.

Application Layer

Presents insights through dashboards, call details, scorecards, and reporting views.

Integration Layer

Moves structured outputs into APIs, exports, CRM workflows, BI tools, and operational systems.

Governance Layer

Defines metric logic, rubric consistency, mapping rules, and data quality assumptions.

Monitoring / Reliability Layer

Tracks processing state, retries, failures, completeness, and confidence handling.

What the Platform Stores

Raw / Source Artifacts

  • - Call records and recording references
  • - Provider/source callbacks and metadata payloads
  • - Ingestion context used for attribution and routing

Intermediate Artifacts

  • - Timestamped transcripts
  • - Speaker-separated segments and utterance boundaries
  • - Stage-level analysis outputs and processing status

Structured Intelligence Outputs

  • - Findings and entities
  • - Metrics, rubric scores, and KPI-ready values
  • - Summaries, QA/compliance signals, and evidence-linked outputs

Delivery Artifacts

  • - Dashboard and call-detail payloads
  • - API responses and export records
  • - Searchable analytics views for reporting and BI workflows

How Architecture Supports Different Teams

Sales Leaders

Objection analysis, buying-signal visibility, rep coaching, and conversion-pattern review.

QA / Compliance

Rubric scoring, required-language detection, exception management, and review prioritization.

Operations Leaders

Call-quality monitoring, escalation-pattern visibility, first-call-resolution analysis, and workflow bottleneck detection.

Marketing / Voice of Customer

Recurring topic detection, market-objection analysis, language-pattern extraction, and message-resonance gap insights.

Architecture Design Principles

  • - Structured outputs over black-box summaries.
  • - Modular analysis stages with clear stage responsibilities.
  • - KPI definitions tied to documented logic and governance.
  • - Dashboard and API parity for core intelligence outputs.
  • - Export-ready data model for downstream analytics and operations.
  • - Built-in support for review, auditability, and exception workflows.
  • - Extensible design for new metrics, rules, and analysis packs.

Data Outputs by Level

Call-Level Outputs

  • - Transcript and speaker turns
  • - Summary, highlights, and action items
  • - Sentiment, objections, risk, and opportunity signals

Manager-Level Outputs

  • - QA scorecards and coaching moments
  • - Risk flags and exception queues
  • - Team KPI trends across calls and cohorts

System-Level Outputs

  • - API records and machine-readable result bundles
  • - CSV/JSON export artifacts for analytics systems
  • - Aggregated dashboard views and integration payloads

Delivery parity matters: the same structured records support UI views, API consumption patterns, and export workflows so operational decisions remain consistent across teams and tools.

Architecture References