Skip to content

Prerequisites and Architecture

This chapter helps platform operators understand what needs to exist (infrastructure and integrations) before enabling Conversation Analytics in a partner-hosted / multi-tenant deployment.

It also provides an operator-level architecture view so you know which subsystem to check when something breaks.

Reference architecture (operator view)

At a conceptual level:

Conversation sources (voice + text)
        │
        ▼
 Ingestion / Connectors
        │
        ▼
 Raw content + metadata storage
        │
        ├──► Transcription (voice) ──► Transcript
        │
        └──► Normalization (text) ──► Text thread
                    │
                    ▼
             AI Assistant job
          (runs enabled AI Tasks)
                    │
                    ▼
          Custom Fields (insight values)
                    │
                    ▼
      Dashboards • Search • Conversation details

Core platform components (what to identify in your environment)

Data ingestion (per channel)

  • Voice call ingestion (audio and metadata)
  • Chat/email/ticket ingestion (message thread + metadata)
  • Normalization/mapping layer (IDs, timestamps, participants, channel types)

Transcription (voice channels)

  • Transcription engine(s)
  • Language strategy (auto-detect vs configured languages)
  • Transcript storage and indexing

AI analysis

  • AI engines: LLM providers/models and their credentials/config
  • AI Tasks: global task library + tenant activation + overrides
  • AI Assistant job: continuous pipeline executing tasks and persisting results

Data storage and retrieval

  • Operational database (tenants, tasks, fields, mapping, status)
  • Search index (conversations, transcripts/threads, custom fields)
  • Analytics storage (for dashboards/aggregations), if separate from operational DB

Observability and governance

  • Logs/metrics/traces for ingestion, transcription, jobs, task executions
  • Audit logs for configuration changes (fields/tasks/overrides)
  • Usage accounting (LLM spend, requests, tokens) per tenant (recommended)

Environment prerequisites (operator checklist)

Networking & security

  • Outbound connectivity to:
  • transcription provider(s) (if external)
  • LLM provider(s) (if external)
  • TLS certificates and secure endpoints for ingestion
  • Secrets management for provider credentials

Compute and scaling

  • Worker capacity for:
  • ingestion processing
  • transcription throughput
  • AI Assistant job concurrency
  • Backpressure mechanisms (queues) to handle spikes and retries

Storage & retention

  • Storage for raw audio (voice) and text threads (omni-channel)
  • Retention policy per tenant (recommended to support configurable retention)
  • Backup and restore strategy

Operational readiness checklist

Before enabling customer tenants, verify:

  • ✅ You can ingest and store conversations for at least one test tenant
  • ✅ You can produce transcripts (voice) or normalized threads (text)
  • ✅ You can run the AI Assistant job and persist Custom Field outputs
  • ✅ You can view outputs in conversation details, dashboards, and search
  • ✅ You have monitoring + alerting for each pipeline stage

Implementation notes

MiaRec uses a queue-based worker model for processing. Key tuning knobs include:

  • Worker concurrency settings
  • Retry policies and dead-letter queues
  • Job schedule and data source configuration

Storage and indexing are handled by the MiaRec platform. Retention controls can be configured per tenant.

Contact MiaRec for detailed capacity planning guidance based on your expected workload (calls/day, average transcript length, tasks enabled).