{"id":8381,"date":"2026-04-28T11:49:01","date_gmt":"2026-04-28T06:19:01","guid":{"rendered":"https:\/\/blogs.infosys.com\/digital-experience\/?p=8381"},"modified":"2026-04-28T12:15:00","modified_gmt":"2026-04-28T06:45:00","slug":"a-pov-on-kermit-conversational-ai","status":"publish","type":"post","link":"https:\/\/blogs.infosys.com\/digital-experience\/digital-commerce\/a-pov-on-kermit-conversational-ai.html","title":{"rendered":"Designing Kermit Through a Multi\u2011Agent Conversational AI Lens"},"content":{"rendered":"<h3><strong>Introduction<\/strong><\/h3>\n<p>This blog walks through the end\u2011to\u2011end architecture of Kermit, explaining how a modern agentic AI platform, cloud\u2011native services, and SAP Commerce integrations can come together to deliver secure, scalable, and intelligent customer interactions.<\/p>\n<h3><strong>What is Kermit?<\/strong><\/h3>\n<p>Kermit is a conversational AI assistant embedded within the client\u2019s B2B Mobile Application. It enables customers to:<\/p>\n<ul>\n<li>Ask questions about products and pricing<\/li>\n<li>Place and manage orders using natural language<\/li>\n<li>Receive personalized recommendations<\/li>\n<li>Access account, property, and order\u2011related information<\/li>\n<li>Kermit has been designed with a code\u2011first, multi\u2011agent architecture and is fully deployed on AWS, ensuring reliability, scale, and strong governance from day one.<\/li>\n<\/ul>\n<h3><strong>Business Use Cases Covered by Kermit<\/strong><\/h3>\n<p>The Kermit Conversational AI platform supports a broad set of business and operational use cases across the digital ecosystem.\u00a0 Below are few of the key use cases:<\/p>\n<p><strong>Customer Self\u2011Service &amp; Support<\/strong><\/p>\n<ul>\n<li>Order status inquiries (by order number, phone number, or account context)<\/li>\n<li>Order cancellation and modification requests<\/li>\n<li>Repair status checks and repair initiation<\/li>\n<li>Account management (profile details, properties, preferences)<\/li>\n<li>FAQs and policy\u2011driven questions (returns, terms &amp; conditions, promotions)<\/li>\n<\/ul>\n<p><strong>Product Discovery &amp; Commerce Enablement<\/strong><\/p>\n<ul>\n<li>Guided product search using natural\u2011language queries<\/li>\n<li>Product availability, pricing, and packaging inquiries<\/li>\n<li>Personalized product recommendations<\/li>\n<li>Promotion and offer discovery<\/li>\n<li>Cart creation and pre\u2011checkout assistance<\/li>\n<\/ul>\n<p><strong>Analytics, Insights &amp; Optimization<\/strong><\/p>\n<ul>\n<li>Identification of order drop\u2011offs and escalations<\/li>\n<li>Measurement of bot\u2011assisted versus human\u2011assisted resolutions<\/li>\n<\/ul>\n<p><strong>Knowledge &amp; GenAI\u2011Driven Assistance<\/strong><\/p>\n<ul>\n<li>Knowledge Q&amp;A using curated enterprise content (FAQs, product catalogs, policies)<\/li>\n<li>Context\u2011aware response generation using LLMs with enterprise guardrails<\/li>\n<li>AI\u2011generated conversation summaries for review and analysis<\/li>\n<\/ul>\n<h3><strong>Architectural Overview<\/strong><\/h3>\n<p>Below represents inferred reference architecture aligning Kermit components with enterprise CX and AI patterns. The inferred architecture assumes Claude Haiku\u202f4.5 on Amazon Bedrock as the underlying LLM for Kermit.<\/p>\n<ol>\n<li>Mobile Front End (iOS &amp; Android Applications)<\/li>\n<li>Security &amp; Identity Layer (Microsoft ADFS, JWT)<\/li>\n<li>Network &amp; Load Balancing Layer (NLB + ALB)<\/li>\n<li>Agentic AI Layer (Orchestrator + Sub\u2011Agents)<\/li>\n<li>Data &amp; Caching Layer (Redis, DynamoDB)<\/li>\n<li>SAP Commerce &amp; Enterprise Integrations<\/li>\n<li>Monitoring, QA, and Analytics Layer<\/li>\n<\/ol>\n<p>Each layer is independently scalable and designed following cloud\u2011native and zero\u2011trust principles.<\/p>\n<h3><strong>Front End:\u00a0 Mobile Application<\/strong><\/h3>\n<p>The Mobile Application is the primary entry point for Kermit and is built as a fully native mobile experience:<\/p>\n<ul>\n<li><strong>iOS:<\/strong> Swift &amp; SwiftUI<\/li>\n<li><strong>Android:<\/strong> Kotlin &amp; Jetpack Compose<\/li>\n<\/ul>\n<p>The chat interface communicates with backend services using <strong>Server\u2011Sent Events (SSE)<\/strong>, enabling real\u2011time streaming responses from the AI in small chunks. This creates a responsive, chat\u2011like experience even when responses involve multiple backend calls or reasoning steps.<\/p>\n<h3><strong>Security &amp; Identity Management<\/strong><\/h3>\n<p>Security is enforced and can be aligned with the respective client&#8217;s enterprise standards:<\/p>\n<ul>\n<li>Microsoft ADFS acts as the centralized Identity Provider<\/li>\n<li>Users authenticate once via the mobile application<\/li>\n<li>JWT tokens are issued post\u2011authentication<\/li>\n<li>Each chat request carries:\n<ul>\n<li>An Authorization Token (standard ADFS access token)<\/li>\n<li>An X\u2011Client Token carrying customer and contextual data<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>Kermit itself never issues authentication tokens. Instead, it validates tokens with the client\u2019s mobile platform, maintaining strict separation of concerns.<\/p>\n<h3><strong>Network &amp; Load Balancing<\/strong><\/h3>\n<p>All inbound traffic flows securely through:<\/p>\n<ol>\n<li><strong>Network Load Balancer (NLB)<\/strong> \u2013 external entry point hosted in the client\u2019s AWS account<\/li>\n<li><strong>Application Load Balancer (ALB)<\/strong> \u2013 internal routing within a private VPC<\/li>\n<\/ol>\n<p>From here, requests are routed to containerized services running on Amazon ECS with AWS Fargate. All components operate inside a private VPC, with AWS VPC Endpoints providing secure, private access to services such as Bedrock, S3, and CloudWatch.<\/p>\n<h3><strong>Prefetch Service: Reducing Latency from First Interaction<\/strong><\/h3>\n<p>To improve perceived performance, Kermit introduces a Prefetch Service:<\/p>\n<ul>\n<li>Implemented as an AWS Lambda function<\/li>\n<li>Triggered immediately after user login<\/li>\n<li>Pre\u2011loads:\n<ul>\n<li>User\u2011specific SAP Commerce data<\/li>\n<li>Frequently accessed product information<\/li>\n<\/ul>\n<\/li>\n<\/ul>\n<p>This proactive cache warming significantly reduces cold\u2011start latency during the user\u2019s first conversational interaction.<\/p>\n<h3><strong>The Agentic AI Layer: Where Intelligence Lives<\/strong><\/h3>\n<h4><strong>Orchestrator<\/strong><\/h4>\n<p>At the core of Kermit is a stateless Orchestrator built using:<\/p>\n<ul>\n<li>FastAPI (Python 3.13)<\/li>\n<li>Uvicorn ASGI server<\/li>\n<\/ul>\n<p>Responsibilities include:<\/p>\n<ul>\n<li>Managing chat sessions<\/li>\n<li>Interpreting user intent using Claude Haiku 4.5 via Amazon Bedrock<\/li>\n<li>Routing requests for specialized sub\u2011agents<\/li>\n<li>Streaming responses back to the client via SSE<\/li>\n<li>Enforcing rate limits and health checks<\/li>\n<\/ul>\n<p>All conversation state persisted externally in DynamoDB, keeping the orchestrator horizontally scalable.<\/p>\n<h4><strong>Sub\u2011Agents (Agent Core)<\/strong><\/h4>\n<p>Kermit uses multiple domain\u2011specific AI agents, each deployed as a separate ECS Fargate task and running on Amazon Bedrock Agent Core using the Strands Agent Framework:<\/p>\n<ul>\n<li style=\"list-style-type: none\">\n<ol>\n<li><strong>Account Info Agent \u2013<\/strong> customer profiles, properties, contacts<\/li>\n<li><strong>Order Agent \u2013<\/strong> full order placement workflows<\/li>\n<li><strong>Order &amp; Recommendations Agent \u2013<\/strong> order history and recommendations<\/li>\n<li><strong>Product Info Agent \u2013<\/strong> catalogue, pricing, fuzzy search<\/li>\n<\/ol>\n<\/li>\n<\/ul>\n<p>Agents communicate securely using SigV4\u2011signed HTTP requests and coordinate via an Agent\u2011to\u2011Agent (A2A) protocol.<\/p>\n<h4><strong>Bedrock, Guardrails, and RAG<\/strong><\/h4>\n<p>The conversational reasoning across Kermit is powered by Claude Haiku 4.5 via Amazon Bedrock.<\/p>\n<p>Key capabilities include:<\/p>\n<ul>\n<li>Bedrock Guardrails to enforce content safety and fertilizer\u2011domain relevance<\/li>\n<li>Retrieval\u2011Augmented Generation (RAG) using Amazon Bedrock Knowledge Bases<\/li>\n<li>Source content stored in Amazon S3 (PDFs, CSVs, structured documents)<\/li>\n<\/ul>\n<p>This ensures responses are both accurate and grounded in client\u2011approved data.<\/p>\n<h4><strong>Data Layer &amp; Caching Strategy<\/strong><\/h4>\n<p><strong>Caching:<\/strong><\/p>\n<ul>\n<li>Amazon ElastiCache Serverless (Redis) is used for caching.<\/li>\n<li>A dedicated Cache Manager Lambda controls read\/write\/invalidation logic<\/li>\n<\/ul>\n<p><strong>Cache tiers include:<\/strong><\/p>\n<ul>\n<li>User\u2011specific SAP Commerce data (TTL: 1 hour)<\/li>\n<li>Global product &amp; pricing data (TTL: 24 hours, refreshed every 4 hours)<\/li>\n<\/ul>\n<p>Transactional operations (order placement, cart submission) always bypass the cache to guarantee data consistency.<\/p>\n<p><strong>Conversation History<\/strong><\/p>\n<p>All conversational states are stored in Amazon DynamoDB:<\/p>\n<ul>\n<li>Orchestrator chat history<\/li>\n<li>Sub\u2011agent Conversation<\/li>\n<li>Session validity enforced within an 8\u2011hour window<\/li>\n<\/ul>\n<h3><strong>SAP Commerce &amp; Enterprise Integrations<\/strong><\/h3>\n<p>Kermit integrates deeply with enterprise systems via:<\/p>\n<ul>\n<li>SAP Commerce APIs exposed through SAP API Management<\/li>\n<li>B2B Mobile Application APIs for business operations<\/li>\n<li>ArcGIS mapping integration for geo\u2011spatial and property\u2011based context<\/li>\n<\/ul>\n<p>This allows the AI to operate with full business awareness rather than acting as a standalone chatbot.<\/p>\n<h3><strong>Observability, QA, and Analytics<\/strong><\/h3>\n<p><strong>Monitoring<\/strong><\/p>\n<ul>\n<li>Amazon CloudWatch for logs and metrics<\/li>\n<li>Centralized error processing using Lambda, DynamoDB, Event Bridge, and SES<\/li>\n<li>Hourly error digests sent to engineering teams<\/li>\n<\/ul>\n<p><strong>Quality &amp; Evaluation<\/strong><\/p>\n<ul>\n<li>Automated LLM evaluations using DeepEval<\/li>\n<li>Deployment\u2011triggered regression testing<\/li>\n<li>Dashboards via Bedrock Analytics and Amazon QuickSight<\/li>\n<\/ul>\n<h3><strong>Deployment Model<\/strong><\/h3>\n<p>All components can be deployed via CloudFormation across:<\/p>\n<ul>\n<li>Dev<\/li>\n<li>QA<\/li>\n<li>Production<\/li>\n<\/ul>\n<p>Compute choices are optimized per workload:<\/p>\n<ul>\n<li><strong>ECS Fargate \u2013<\/strong> APIs and AI agents<\/li>\n<li><strong>AWS Lambda \u2013<\/strong> prefetch, caching, error handling<\/li>\n<\/ul>\n<h3>Closing Thoughts<\/h3>\n<p>Kermit represents a shift from traditional rule\u2011based digital assistants to a true agentic AI platform\u2014secure, scalable, and deeply integrated with enterprise commerce systems.<\/p>\n<p>By combining multi\u2011agent orchestration, modern LLMs, and cloud\u2011native architecture, one of our clients has laid the foundation for intelligent, context\u2011aware customer engagement in the agriculture sector\u2014with clear pathways for future expansion into voice and advanced multimodal experiences.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Introduction This blog walks through the end\u2011to\u2011end architecture of Kermit, explaining how a modern [&hellip;]<\/p>\n","protected":false},"author":580,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"footnotes":""},"categories":[499,324],"tags":[],"coauthors":[497],"class_list":["post-8381","post","type-post","status-publish","format-standard","hentry","category-artificial-intelligence","category-digital-commerce"],"acf":[],"_links":{"self":[{"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/posts\/8381","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/users\/580"}],"replies":[{"embeddable":true,"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/comments?post=8381"}],"version-history":[{"count":10,"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/posts\/8381\/revisions"}],"predecessor-version":[{"id":8416,"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/posts\/8381\/revisions\/8416"}],"wp:attachment":[{"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/media?parent=8381"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/categories?post=8381"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/tags?post=8381"},{"taxonomy":"author","embeddable":true,"href":"https:\/\/blogs.infosys.com\/digital-experience\/wp-json\/wp\/v2\/coauthors?post=8381"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}