← Back
Google Research

How Google Duplex evolved to also call businesses for you in Gemini

•7 min read•
#agents#llm
How Google Duplex evolved to also call businesses for you in Gemini
Level:Advanced
For:AI Engineers, Product Engineers
✦TL;DR

Google Duplex, originally launched in 2018, has been re‑architected to run on the Gemini foundation model stack, enabling the new “Tasks in Gemini” app to autonomously call businesses on behalf of users. The evolution moved from early closed‑domain recurrent networks to a unified Gemini‑based architecture, and the system now powers on‑device features like Pixel Call Assist and large‑scale verification that has appeared in Search and Maps over one trillion times. This unified, low‑latency stack combines speech understanding, contextual reasoning, and speech synthesis, allowing engineers to embed agentic phone‑calling capabilities directly into consumer products.

⚡ Key Takeaways

  • Business information sourced from Duplex phone calls has been shown in Search and Maps over one trillion times.
  • The current architecture is a unified stack built entirely with the Gemini foundation model, replacing earlier bespoke recurrent pipelines.
  • The system runs on‑device across five consecutive Pixel generations, supporting features such as Call Screen and Hold for Me with low latency.
  • Integration is exposed through the “Tasks in Gemini” app, which acts as a personal AI collaborator for real‑world chores like business calls.
  • The architecture requires high‑quality speech understanding, contextual reasoning, and speech synthesis to operate within sub‑second turn‑taking constraints.

🔧 Tools & Libraries

GeminiGoogle DuplexPixel Call Assist
💡 Why It Matters

Engineers can now leverage a production‑ready, on‑device Gemini‑based stack to add autonomous, real‑time phone‑calling agents to apps, reducing the need to build separate speech and reasoning pipelines and dramatically scaling verification workflows.

✅ Practical Steps

  1. Use the “Tasks in Gemini” app API to submit a task request (e.g., “call a hair salon”) and receive the completed call outcome.
  2. Deploy the Gemini‑based conversational stack on supported Pixel devices to enable on‑device call assistance features like Call Screen.
  3. Incorporate the unified speech‑understanding, reasoning, and synthesis pipeline into existing Google Search or Maps integrations to automate merchant verification.

Want the full story? Read the original article.

Read on Google Research ↗

More like this

Censored vs. Truncated Regression in Python, Explained

Towards Data Science•#llm

New in AWS Quick: More capable flows, an updated web experience, and agents that work together

AWS ML Blog•#agents

Q&A: Enabling more human-like interactions and understanding in multimodal foundation models

MIT News AI•#llm

Welcome Inkling 2 Mini, Cutting-edge Performance for your Local Setup

Hugging Face Blog•#llm

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING