ctxdexSynthetic model data for learning
Customer Experience

Realtime Voice Support Agent

Combines speech processing, support reasoning and enterprise context for a realtime customer interaction.

Explores text-orchestrated and native realtime approaches to voice support while keeping audio input, audio output and text-model usage distinct.

VoiceRealtimeAudioCustomer Support
Explore Scenario Economics →

Design variants

The same solution, implemented 2 ways. Choosing one changes the workflow and dependencies described below; it does not calculate anything.

How it works

6 operations and 11 activities, in the order they run. Expand an operation to see its activities.

  1. Capture the customer's speech and turn it into a request the support pipeline can read.

  2. Interpret the request and plan what the answer needs.

  3. Find and assemble the enterprise knowledge the answer is grounded in.

  4. Look up whatever operational record the request needs and normalize it.

  5. Write the grounded answer as text.

  6. Speak the answer back to the customer.

What this solution uses

For Text-Orchestrated Voice. An activity that calls a model and a service appears in both groups.

AI / model work6

  • Speech 1 RecognitionTranscribe / Recognize Customer SpeechDefault: Ctxdex Speech 1
  • Model PlanningInterpret and Plan RequestDefault: Ctxdex Intelligence 4 Fast
  • Embedding ModelEmbed QueryDefault: Ctxdex Retrieval 3 Embed
  • Reranking ModelRerank CandidatesDefault: Ctxdex Retrieval 2 Rerank
  • Model GenerationGenerate Grounded ResponseDefault: Ctxdex Intelligence 4 Fast
  • Speech 2 Realtime SynthesisSynthesize Spoken ResponseDefault: Ctxdex Speech 2 Realtime

Services0

None in this design

    Application / workflow work5

    • Deterministic Query PreparationPrepare Retrieval Query
    • Local Enterprise SearchSearch Enterprise Knowledge
    • Deterministic Context AssemblyAssemble Grounded Context
    • External Tool / Application LogicExecute Operational Tool
    • Deterministic NormalizationNormalize Operational Result

    Design details

    Business unit
    Customer Voice Session
    Design variants
    2
    Operations
    6
    Activities
    11
    Model-backed activities
    6
    Priced service dependencies
    0
    Application / workflow activities
    5

    Explore Scenario Economics

    Scenario Economics calculates this solution. There you can choose a compatible model or offering for each activity, set workload assumptions and volume, and change the other inputs Text-Orchestrated Voice supports.

    Open Scenario Economics →