Leo Choi, Director of Engineering at SleekFlow, explains how their AgentFlow platform leverages Microsoft Azure AI technologies and modern cloud architecture to enable responsive, scalable, and secure enterprise customer conversations.

How SleekFlow Uses Azure and AI to Orchestrate Enterprise Customer Conversations

Introduction

SleekFlow, founded in 2019, supports enterprises in transforming almost-lost customer moments into opportunities for connection and growth. Their AgentFlow conversational AI platform orchestrates over 600,000 daily interactions across channels like WhatsApp, Instagram, web chat, and email.

Addressing the Challenge of Conversational AI at Scale

Enterprises demand intelligent, always-available conversations. Initial challenges included poor scalability, vector search delays, and high operating costs, especially during peak and multi-tenant workloads. These challenges hindered both development and customer onboarding, signaling the need for an enterprise-grade technical foundation.

Azure as the Core Platform

SleekFlow chose Microsoft Azure to power AgentFlow, moving beyond basic compute to a full cloud ecosystem able to coordinate specialized AI agents and scale globally. Key Azure components include:

Azure Cosmos DB

  • Serves as memory/context backbone, handling short- and long-term interactions, chat histories, with vector embeddings.
  • Enables low-latency (15–20 ms) responses essential for real-time conversations.

Azure AI Foundry + OpenAI Models

  • Drives multilingual natural language generation and understanding.
  • Allows conversation agents to operate natively in languages like English, Chinese, and Portuguese.

Semantic Kernel

  • Orchestrates interactions among specialized agents pulling knowledge from context, transactional data, and vector embeddings.
  • Example: Pricing agent retrieves latest data, summarizer agent prepares content, coordination agent handles human handoff.

Azure Container Apps & KEDA Autoscaling

  • Ensures elastic scaling from a few idle containers to 160+ on demand.
  • Utilizes the Pay-As-You-Go model for cost efficiency.
  • Supports event-driven spikes (e.g., promotional campaigns) with stable performance.

DiskANN and Microsoft Orleans

  • DiskANN powers high-speed vector search (semantic lookups in ~15-20 ms).
  • Orleans manages in-memory clustering for seamless conversational flow.

Operational Benefits Achieved

  • Retrieval-augmented generation recall improved from 50% to 70%.
  • End-to-end execution speed is 50% faster.
  • Real-time qualification of leads, cart recovery, and accelerated support resolution.

Security and Compliance First

  • Azure Cosmos DB enforces tenant isolation, RBAC, encryption, and auditable data partitions.
  • All AI models (Azure OpenAI, Microsoft Phi) are deployed in SleekFlow’s private Azure environment, guaranteeing strict data privacy and compliance.
  • Azure certifications (ISO 27001, SOC 2, GDPR) and features like regional residency support global enterprise trust requirements.
  • Human override is always possible, with workflows orchestrated via Azure Container Apps and App Service.

Impact on Development Velocity

  • Migration from monolith to microservices on Azure Container Apps cut release cycles from 5 weeks to near-daily, with blue-green deployments enabling zero-downtime updates.
  • Reduced latency allows engineers to focus on feature development instead of constant debugging.

Conclusion & Future

SleekFlow’s journey illustrates how tightly-coupled Azure AI, cloud-native, and orchestration tools can drive both business outcomes and engineering agility. Their experience provides a model for building enterprise-ready agentic AI solutions focused equally on speed, scalability, compliance, and human-like customer engagement.

This post appeared first on “Microsoft Tech Community”. Read the entire article here