Static IVRs to Agentic Voice AI: Building Real-Time Intelligent Conversations

Static IVRs to Agentic Voice AI: Building Real-Time Intelligent Conversations

by Pablo Salvador Lopez

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQLNatural language and machine translation
🎙 Pablo Salvador Lopez 👥 5K 📅 October 23, 2025 ⏱ 23 min 👁 47 📄 expert opinion 🧭 2026-08-15
Available in: English (current) Français

Keywords

IVRvoice agentsspeech-to-texttext-to-speechorchestration

Summary

Pablo Salvador Lopez, Principal AI Application Development Architect at Microsoft, presents a session from MLOps World 2025 on building real-time agentic voice AI systems. He begins by highlighting the limitations of static IVRs and the common frustration of users wanting to speak to a human. He then introduces an accelerator built by his team that integrates three layers: AI, application, and telephony. The AI layer involves choosing between pre-built speech-to-speech APIs or building a custom pipeline with speech-to-text, LLM/SLM reasoning, and text-to-speech. He emphasizes the importance of turn detection and interruption cancellation for natural conversations. A live demo shows a multi-agent system handling an insurance claim, with agents for authentication, policy questions, and claim intake, seamlessly transferring between each other and switching languages. The application layer discusses orchestration frameworks, scaling, and protocol choices (WebSocket vs WebRTC). The telephony layer uses Azure Communication Services (ACS) to connect to phone networks and integrate with existing IVR systems. The talk concludes with an announcement of a new ‘Boys Live API’ and an invitation to a deeper technical session. The speaker stresses that everything is open-source and framework-agnostic.

184 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable insights into the practical implementation of agentic voice AI, drawing from real-world experience at Microsoft. The live demo effectively illustrates the capabilities and potential of such systems, showcasing multi-agent collaboration, language switching, and integration with telephony. The argumentation is persuasive, emphasizing the shift from rigid IVRs to flexible, context-aware agents. However, the technical depth is limited, with many concepts mentioned but not fully explained. The speaker acknowledges skipping details due to time constraints, which may leave some audience members wanting more. The value lies in the high-level architecture and the demonstration of what is achievable, rather than in-depth technical guidance.

Scientific Rigor, Source Quality, Title Accuracy

The talk is based on the speaker’s professional experience and a live demo, but it lacks explicit citations to scientific literature or external sources. The only source mentioned is the MLOps World conference website, which is not a scientific reference. The title accurately reflects the content, focusing on the evolution from static IVRs to agentic voice AI. The presentation is more of an expert opinion and industry showcase than a rigorous scientific exposition. The speaker mentions that the code is open-source but does not provide specific links, limiting the verifiability of the claims. Overall, the scientific rigor is moderate, with a strong practical orientation but limited academic grounding.

227 words

Title / Content Match

The title accurately reflects the content, which focuses on transitioning from static IVRs to agentic voice AI.

Quality & Reliability

7/10

The talk is based on practical experience from a Microsoft principal architect, with a live demo. However, it is largely anecdotal and lacks detailed technical depth or citations. The speaker mentions open-source availability but provides no specific links or references.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

  • MLOps World — Conference website where the talk was recorded.

Concurring Sources

  • Azure Communication Services — Microsoft's service for telephony integration, mentioned in the talk.

Contribution & Novelties

The talk presents a practical, end-to-end architecture for building agentic voice AI systems, integrating AI, application, and telephony layers into a single stack. It demonstrates a working multi-agent system with real-time voice interaction, including language switching and handoffs between specialized agents. The speaker emphasizes that the solution is open-source and framework-agnostic, offering a blueprint for organizations to build similar systems. The main novelty is the integration of these components into a cohesive accelerator, with a focus on production readiness and scalability.

Pour aller plus loin :

  • Azure Communication Services — Official documentation for the telephony service used in the demo.
  • WebRTC — Standard for real-time communication, relevant to the protocol discussion.
  • Agentic AI — Overview of agentic AI concepts, though the term is not yet widely established.

127 words

Radar Profile

The radar profile shows a balanced but moderate performance across all dimensions, with slightly higher scores in information quality and technical level, reflecting the practical expertise of the speaker. The lower score in information quantity suggests that the talk, while insightful, does not provide exhaustive detail.

Reliability 6/10

💬 No comments were provided for analysis.