Skip to main content
KaBuM! logo

CUSTOMER STORY

KaBuM! powers AI Commerce with Genie Agents, Lakebase Postgres and Unity Gateway on Databricks

KaBuM! powers AI Commerce with Genie Agents, Lakebase Postgres and Unity Gateway on Databricks

Hours to minutes for business reports

Genie Agents absorbed data team requests, giving business users accurate reports and deeper analysis directly

3 months to launch AI Commerce

Lakebase Postgres low-latency memory and Unity Gateway model management sped development of the “Compre com o Ninja” shopping agent

Dozens of use cases in production

Apps, agents and data pipelines run on Databricks, with Unity Gateway centralizing LLM access, costs and guardrails

Pablo Henrique de Freitas

Databricks helped KaBuM! accelerate innovation and democratize access to data and AI, transforming complex analyses into agile decisions.

- Pablo Henrique de Freitas, Data and AI Manager, KaBuM!

KaBuM! is the largest technology and gaming e-commerce platform in Latin America. Its mission is to connect, transform and simplify lives through technology. It does so by providing products, services and knowledge with enhanced operational efficiency. As retail rapidly evolves toward AI Commerce and agentic systems, KaBuM! required an agile foundation. It needed to serve business users and power real-time customer applications. Modernizing on GCP with the Databricks Data + AI Platform, KaBuM! standardized its architecture on Lakebase Postgres, Genie Agents, Unity Gateway and Unity Catalog. Today, leadership can access real-time insights within minutes, while AI engineers can rapidly launch performant conversational agents like "Compre com o Ninja".

Addressing data bottlenecks and latency in AI Commerce

The retail sector is undergoing a major transformation driven by innovations such as AI Commerce. To stay at the forefront of the e-commerce sector, KaBuM! needed to provide operational efficiency across internal leadership decision-making and customer-facing experiences. However, internal stakeholders faced significant friction in quickly obtaining business data and insights. Obtaining reports followed a traditional request process. Business users opened tickets with the central data and AI team, requiring engineers to build queries, views and static dashboards manually. This process created dependency on service SLAs and slowed strategic decision-making.

At the same time, KaBuM! set out to build a standardized internal framework for custom AI agent development. A primary component of this framework is the agent memory layer. For interactive customer-facing AI agents, short-term conversational memory must be read and written during every customer interaction. Operating in a latency-sensitive environment meant that any slowness in retrieving memory or managing large language model (LLM) endpoints directly degraded the user experience. KaBuM! needed a consolidated platform capable of democratizing data access internally while providing low-latency storage and centralized LLM governance.

"There was a pain point in obtaining data, information and insights in a more agile and fast manner," said Vitor Teixeira, Planning, Data and AI Director at KaBuM!. "This demand followed the traditional paths existing in most companies: opening a ticket for the central team to build queries and fulfill the request. We needed a solution that could deliver answers in minutes."

Implementing a standardized AI agent framework with Databricks

To modernize its architecture, KaBuM! implemented the Google Cloud-hosted Databricks Data + AI Platform. To solve the data request bottleneck, KaBuM! introduced Genie Agents.

KaBuM! began with a staged wave rollout. It started with the CEO and directors, then expanded to department heads, managers and key business users. KaBuM! deployed two core agents in production.

General business agent deployment

The first is a general business agent connected to KaBuM!’s key business indicators, including sales revenue, margins, price indices, inventory volume and marketing data such as website visits and media investment. Business teams now interact directly with data in natural language. This enables real-time exploration during executive meetings and helps leaders move from static dashboards to dynamic, on-demand analysis.

Specialized marketing agent for deep queries

The second is a specialized marketing agent designed for deep domain queries. Together, these agents standardize how teams discover insights. They reduce dependency on manual queries while keeping business users close to trusted data during day-to-day decisions.

Building a standardized AI agent development framework

For customer-facing AI applications built using LangGraph, KaBuM! integrated Lakebase Postgres, Unity Catalog and Unity Gateway directly into its standard AI agent framework. KaBuM! evaluated various market databases and selected Lakebase as the default short-term and long-term memory layer due to its competitive pricing and deep integration within the Databricks Platform. Lakebase handles the high-frequency reads and writes required by agent memory. Over a recent 10-day period, it processed more than 413,000 transactions with only 8 rollbacks, averaging about 0.71 milliseconds per transaction. A 99.85% cache hit rate and zero deadlocks or lock waits ensure consistent memory access during live customer conversations.

To manage model interactions and governance, KaBuM!’s 12 AI engineers route LLM traffic through Unity Gateway alongside Unity Catalog. Unity Gateway provides a single control point for LLM endpoints and governance. It allows KaBuM! to route complex tasks to larger models and less complex tasks to smaller models within the same workflow, eliminating the need to manage disparate API keys from multiple providers. Security guardrails, interaction tracking and token consumption monitoring are all managed centrally.

"Unity Gateway gave us the flexibility to use different types and sizes of models centralized in a single place without needing multiple API keys," explained João Marcos Matos, Senior AI Engineer at KaBuM!. "Lakebase serves as our default memory layer because it is fast, precise and fully integrated into our data platform."

Accelerating insights and launching AI Commerce rapidly

Today, KaBuM! runs dozens of use cases in production on Databricks, spanning apps, agents and data pipelines. Deploying Databricks across data and AI workflows delivered immediate operational efficiency and accelerated innovation for KaBuM!. Genie Agents democratized data access, transforming internal meetings from static slide reviews into real-time data exploration. Commercial managers now identify trends and adjust promotional and campaign strategies in minutes rather than waiting hours or days for ticket resolution.

For AI engineering, standardizing on Lakebase Postgres, Unity Catalog and Unity Gateway improved developer productivity. This unified foundation allowed KaBuM! to develop and launch "Compre com o Ninja" (AI Commerce). The result is an innovative, personalized conversational shopping experience for gaming and tech enthusiasts, delivered in just three months.

By leveraging Databricks, KaBuM! achieved key performance improvements: 

  • Accelerated report generation: Response times for business analyses dropped from hours to minutes, reducing data team ticket volume. 

  • Faster time to market: The "Compre com o Ninja" AI Commerce experience was launched in approximately three months. 

  • Centralized cost and model control: Streamlined token tracking, security guardrails and interaction monitoring for 12 AI engineers across multiple model providers via Unity Gateway and Unity Catalog. 

  • Fast agent memory: Lakebase averaged about 0.71 milliseconds per transaction across more than 413,000 transactions, with a 99.85% cache hit rate.

"Databricks will continue to be our central data and AI platform, allowing KaBuM! to remain at the forefront of retail and create increasingly personalized experiences for our customers," concluded Pablo Henrique de Freitas, Data and AI Manager at KaBuM!. Looking ahead, KaBuM! plans to roll out specialized Genie Agents to tactical and operational teams and expand Lakebase as the default memory layer for all new AI agents.

KaBuM! on Databricks Genie Agents, Lakebase Postgres and Unity Catalog

KaBuM! uses the Databricks Data + AI Platform to democratize business data access, speed up strategic decision-making and power innovative AI Commerce customer experiences.