Nandhu S KumarAI & MACHINE LEARNING
← PROJECT INDEXCASE STUDY / 02

AI Voice Agent

A phone agent built with Gemini Live and Twilio that follows call objectives and uses configured tools.

DISCIPLINE

AI Agents

TIMELINE

Started May 2026

BUILT WITH

Gemini Live, Twilio, Tool execution

System architectureFIG. 02
CALLTWILIOLIVE MODELGEMINITOOLSCONTROLLEDCONTEXTOBJECTIVE +HISTORYCONVERSATION ↔ ACTION ↔ FEEDBACK
  1. Call

    Twilio provides the phone connection for the conversation.

  2. Live model

    Gemini Live processes the live conversational interaction.

  3. Context & tools

    The workflow maintains objectives and relevant history, validates inputs, and controls tool execution.

  4. Traceability

    Transcripts and workflow outputs support review of the agent’s behavior.

AI AgentsConceptual overview
INSIDE THE SYSTEM / SELECT A STAGE

Twilio provides the phone connection for the conversation.

Problem

A phone agent needs to remember the call objective, follow the conversation, and use tools without losing track of the task.

Implementation

I connected Gemini Live and Twilio, with call objectives, conversation history, tool outputs, and transcripts managed throughout each call. Configuration and validation control which actions the agent can take.

Technical decisions

  • Integrated Gemini Live with Twilio for live phone conversations.
  • Managed relevant context across objectives, transcripts, and tool responses.
  • Introduced input and output validation and controlled tool execution.
  • Focused workflow optimization on latency, context usage, model efficiency, and traceability.

Results

The application connects phone conversations to tool calls and records transcripts and outputs for review.

NEXT / KEEP EXPLORING

Scientific GPT

Send a message

Contact me

Tell me about the role, project, or question you have in mind.