Zepto Scales Customer Support on Databricks and MLflow

Login

Why Databricks

Discover

For App Developers

For Executives

For Startups

Lakehouse Architecture

Databricks AI Research

Customers

Customer Stories

Partners

Partner OverviewExplore the Databricks partner ecosystem

Partner SpotlightFeatured partner announcements

Partner ProgramExplore benefits, tiers and how to become a partner

Cloud ProvidersDatabricks on AWS, Azure and GCP

Find a PartnerDiscover Databricks partners for your needs

Partner SolutionsFind custom industry and migration solutions

Product

Databricks Data + AI Platform

Platform OverviewA unified platform for data, analytics and AI

Data EngineeringETL and orchestration for batch and streaming data

AI AssistantAgentic coworker for business users

Data WarehousingServerless data warehouse for SQL analytics

Application DevelopmentQuickly build secure data and AI apps

DatabasePostgres for data apps and AI agents

Artificial IntelligenceBuild and deploy ML and GenAI applications

GovernanceUnified governance for all data, analytics and AI assets

Business IntelligenceIntelligent analytics for real-world data

SecurityOpen agentic SIEM built for the AI era

Customer Data PlatformAgentic CDP embedded in Databricks

SharingOpen data sharing for data, analytics and AI

Integrations and Data

MarketplaceOpen marketplace for data, analytics and AI

IDE IntegrationsBuild on the Lakehouse in your favorite IDE

Partner ConnectDiscover and integrate with the Databricks ecosystem

Pricing

Databricks PricingExplore product pricing, DBUs and more

Cost CalculatorEstimate your compute costs on any cloud

Open Source

Open Source TechnologiesLearn more about the innovations behind the platform

Solutions

Databricks for Industries

Telecommunications

Media and Entertainment

Financial Services

Public Sector

Healthcare & Life Sciences

Retail

Manufacturing

See All Industries

Cross Industry Solutions

AI Agents

AI Governance

Cybersecurity

Marketing

Migration & Deployment

Data Migration

Forward Deployed Engineering

Solution Accelerators

Explore AcceleratorsMove faster toward outcomes that matter

Resources

Learning

TrainingDiscover curriculum tailored to your needs

Databricks AcademySign in to the Databricks learning platform

CertificationGain recognition and differentiation

Free EditionLearn professional Data and AI tools for free

University AllianceWant to teach Databricks? See how.

Events

Data + AI Summit

Data + AI World Tour

AI Days

Event Calendar

Blog and Podcasts

Databricks BlogExplore news, product announcements, and more

AI BlogExplore our AI research and engineering work

Data Brew PodcastLet’s talk data!

Get Help

Customer Support

Documentation

Community

Dive Deep

Resource Center

Demo Center

Architecture Center

About

Company

Who We Are

Our Team

Databricks Ventures

Contact Us

Careers

Working at Databricks

Open Jobs

Press

Awards and Recognition

Newsroom

Security and Trust

Security and Trust

Ready to get started?

Get a Demo

Login

Contact Us

Try Databricks

All blogs

/

Engineering

Table of contents

Zepto's Push for Reliable, Real-Time Customer Support

Why "Just Ship the Agent" Breaks at Scale

Why Evaluation Framework and its Outcomes

Building the Framework: A Dual Loop for Confidence and Control

The Composable Architecture Agent Stack

The Framework in Action

What We Learned Running This at Scale

Recommendations for Agent Builders

Conclusion

Table of contents

Table of contents

Zepto's Push for Reliable, Real-Time Customer Support

Why "Just Ship the Agent" Breaks at Scale

Why Evaluation Framework and its Outcomes

Building the Framework: A Dual Loop for Confidence and Control

The Composable Architecture Agent Stack

The Framework in Action

What We Learned Running This at Scale

Recommendations for Agent Builders

Conclusion

Data Science and ML

September 9, 2026

Evaluation-First AI Agents: How Zepto Scales Customer Support on Databricks and MLflow

How Zepto uses Databricks and MLflow to build evaluation-first AI agents that manage 80%+ of support tickets, cut support costs by 65%, and deliver payback in under one month

by Gireesh Sreedhar KP, Deepak Dhankani and Eash Sharma