All works

CASE STUDY / TEG / TICKETEK · PRODUCTION SYSTEM

Stella: a multi-agent ticketing assistant

A conversational assistant connecting event discovery, live pricing, ticket availability, and customer support through three specialised agents.

MY ROLEMachine Learning Engineer
TECHNOLOGIES
Google ADKMulti-agent systemsGCP
Conceptual illustration of connected translucent green nodes around a ticket, representing Stella’s agent orchestration.
AI-generated conceptual illustration.

01 / THE PROBLEM

What needed to work.

Finding an event and deciding whether to book involves several connected questions: what’s on, what tickets are available, what they cost, and where to get help. Stella brings these needs together in a conversational assistant.

02 / MY APPROACH

How I approached it.

01

Coordinate specialised agents

Designed and productionised the assistant with Google ADK, orchestrating three specialised agents across event discovery, real-time pricing, ticket availability, and customer support.

02

Connect the conversation to services

Integrated six tools and five external services so the assistant can work with event and ticketing information as part of the conversation.

03

Support real customer journeys

Deployed the conversational system to support customer sessions at scale, with engagement and purchase conversion reported alongside monthly session volume.

03 / SYSTEM OVERVIEW

Connecting the pieces.

  1. 01Customer question
  2. 02Google ADK orchestration
  3. 03Specialised agents & tools
  4. 04External event services
  5. 05Conversational response

The engineering focus was orchestration: connecting specialised agent behaviour with the tools and services needed to answer a customer’s question. The system spans both discovery and transactional information, rather than treating each as an isolated conversation.

STELLA / SYSTEM ARCHITECTURE01
Multi-agent orchestration3 agents · 6 tools · 5 services

04 / OUTCOMES

Results in context.

50K+

Monthly sessions

Session volume supported by the assistant.

68%

User engagement

Up to 68% engagement, as reported in the résumé.

6.9%

Purchase conversion

Reported conversion for the assistant experience.

Public overview of my contribution at TEG / Ticketek. The workflow is a conceptual summary; internal prompts, service contracts, and implementation details are not reproduced.

Orbit

AI guide to my work

Type or talk to Orbit
A LITTLE CONTEXT, ON DEMAND

What would you like
to know about Bhavya?

Explore the systems he’s built, the problems he’s solved, and the research behind his work.

AI answers from my portfolio. Messages are sent to OpenRouter. Voice input uses your browser’s speech service. Talk to me directly ↗