Customer service in iGaming was hitting a wall
Conduet provides AI-powered customer service for the iGaming industry. Their platform, GameLM, needed to handle thousands of player queries — from bet disputes and account issues to responsible gambling checks — across multiple CRMs and jurisdictions.
But the existing system was tightly coupled, single-tenant and difficult to scale. Onboarding new customers required manual developer intervention. And agents were still doing most of the work manually, with limited tooling to help them resolve queries faster.
They needed an intelligent, scalable platform
Pixie Labs rebuilt GameLM from the ground up, designing a multi-tenant architecture, productionising the AI layer, and delivering two operating modes — Copilot, where AI assists human agents in real time, and Full Automation, where an autonomous agent resolves queries end-to-end without human intervention.
We helped Conduet achieve…
01.
500+ hours saved
In the first month of deployment, GameLM saved over 500 hours of agent time by autonomously resolving queries that previously required manual handling.
02.
38,000 tickets handled
GameLM processed 38,000 customer service tickets in its first month, across both Copilot and Full Automation modes for a major US sportsbook.
03.
Built for multi-tenancy
The platform now supports multiple customers, CRMs and jurisdictions from a single codebase, with config-driven onboarding that replaces weeks of developer setup.
Using our Vital Systems™ process to drive progress
Diagnosing issues
We audited the existing GameLM codebase, infrastructure and integration points. Interviews with Conduet's team and their first customer revealed bottlenecks in scalability, observability and agent workflow — giving us a clear picture of what needed to change.
Planning priorities
We worked with Conduet leadership to define a phased delivery plan. Phase 1A focused on multi-tenancy, Salesforce integration and production-grade infrastructure, with clear milestones and success criteria tied to customer onboarding timelines.
Rebuilding systems
We rebuilt the platform on AWS with a hybrid Rails and Node.js architecture — Rails for orchestration and admin, Node.js for real-time AI request handling. LangGraph powers the agent logic, with LangSmith providing full observability across every conversation.
Priming for growth
With a jurisdictional architecture, pluggable CRM and PAM adapters, and a Snowflake-powered analytics layer, GameLM is now built to onboard new customers and expand into new markets without re-platforming.
The result is an AI-native customer service platform that balances automation with human oversight. With a modular architecture, real-time observability and built-in safety guardrails, Conduet now has a robust foundation to grow across customers, territories and use cases.
01
LangGraph + LangSmith for agent orchestration and observability.
02
Ruby on Rails for orchestration and Node.js for real-time API handling.
03
AWS ECS Fargate, RDS and ElastiCache for scalable infrastructure.
04
Snowflake + dbt for cross-platform analytics and reporting.
05
Salesforce Lightning, Zendesk and Amelco PAM integrations.
06
Kinde for multi-tenant user management and auth.
TODO: Add quote from Conduet