Contextual Agentic Workflow and AWS Marketplace Listing within BOX program
.jpg)
Key Challenges
Perception Grid needed to unify fragmented data sources into a single intelligent metadata layer without duplicating customer data. They also required a scalable agentic AI solution capable of semantic search across multimodal assets, including text, images, videos, and 3D models, while building a secure, serverless AWS architecture that could support future growth and enable a successful AWS Marketplace listing.
Key Results
Ankercloud delivered a contextual agentic workflow powered by AWS, enabling accurate semantic discovery of multimodal assets and intelligent retrieval of contextual information. The solution provided a scalable, serverless architecture, enhanced user experience with rich AI-driven search results, successfully passed the AWS Foundational Technical Review (FTR), and achieved a managed SaaS listing on the AWS Marketplace with automated tenant onboarding and streamlined customer procurement.
Overview
Perception Grid is developing a specialized platform designed to manage, label, and organize a diverse array of digital assets, including 3D models, videos, images, and audio. By leveraging a graph database to establish data relationships, they aim to provide a novel way for users to interact with media and collaborate. In partnership with Ankercloud, they initiated a project to transition from a SaaS discovery phase into a fully functional prototype on AWS.
Challenges
The primary objective was to overcome several technical and functional hurdles:
- Data Fragmentation: The challenge of unifying disparate data sources into a single
intelligent metadata layer on AWS without copying customer data or creating another
data island. - Complex Asset Search: Implementing a semantic search platform that allows users to
navigate through various media formats in a centralized manner. - Multimodal Ingestion: Creating a generic agentic pipeline capable of ingesting and
processing diverse formats like text, images, and complex 3D models.
Solution
Ankercloud developed a Contextual Agentic Workflow using the AWS Strands Framework.
Key components include:
- Agentic Intelligence: A single reasoning agent built using the AWS Strands Framework can handle tool selection, reasoning, and response summarization within a unified workflow. The agent operates through an iterative Plan → Action → Thought cycle, enabling it to decompose complex user requests, select the most relevant tools, evaluate intermediate results, and refine its reasoning before generating a final response. It is capable of reframing user queries to improve semantic understanding and ensure high-accuracy retrieval from multimodal indices containing documents and 3D models.To find the most relevant digital assets, the agent can analyze user intent, generate optimized search queries, retrieve context from the knowledge source, and rank results based on relevance, confidence, and business context. This approach improves retrieval precision, reduces search ambiguity, and delivers more accurate and contextually relevant asset recommendations to end users.
- Custom Tooling: Developed two custom tools for retrieving text chunks and S3 URIs of 3D models (.glb files) from an Amazon OpenSearch Vector Data Base
- Scalable Architecture: The infrastructure is built on AWS using Lambda functions (deployed via Docker containers from ECR), Amazon API Gateway, and Amazon Bedrock for API-driven model access.
- Infrastructure as Code (IaC): Utilized Terraform to automate the deployment of the Lambda service and API Gateway, ensuring a reproducible and scalable environment.
- Monitoring: Integrated Amazon CloudWatch to provide detailed logs and dashboards for monitoring the health and performance of the agentic workflow.
Architecture Diagram
Architecture Description
- API Management: Amazon API Gateway functions as the primary interface connecting diverse client applications to the backend agentic processes.
- Serverless Execution: AWS Lambda, utilizing containerized images from Amazon ECR, orchestrates the internal reasoning and workflow of the agent.
- Container Registry: Amazon ECR serves as a centralized repository for Docker images required by Lambda and Fargate compute environments.
- Vector Storage: An Amazon OpenSearch collection manages multimodal embeddings, facilitating efficient storage and retrieval of text and 3D model data.
- Operational Insight: Amazon CloudWatch provides robust logging capabilities to support system auditing, performance analysis, and health monitoring.
- Foundation Models: Amazon Bedrock provides secure, API-driven access to large language models utilized within the agentic architecture.
Business Outcome
Advanced Semantic Discovery: The system successfully provides detailed context regarding scanned objects and identifies new relationships among assets, such as connections between characters.
Operational Scalability: A working AWS infrastructure was delivered that scales with user growth and remains easily expandable for future AWS service integrations.
Rich User Experience: The platform enables users to receive comprehensive responses that include both textual information and direct S3 URI references to source files like PDFs and 3D models.
Post-Deployment Success: Following the completion of this workload, the solution successfully underwent a Foundational Technical Review (FTR) and the product was officially listed in the AWS Marketplace.
Marketplace listing within BOX program
From a business perspective, listing the product on the AWS Marketplace as a managed SaaS solution involves aligning the service delivery, pricing, and billing processes with AWS standards to simplify procurement for customers. In addition, Ankercloud built an integration component which helped PG to automate their tenant onboarding and make it seamless.
The business-focused steps and considerations include:
- Adopting a Managed Service Model: The product is positioned as a fully hosted SaaS application where the vendor manages all infrastructure, meaning the customer has no resources to manage in their own account.
- Selecting a Strategic Pricing Model: A "Contract with Usage-Based" model was chosen to provide customers with a predictable monthly baseline while ensuring fairness through pay-as-you-go billing for extra consumption.
- Establishing Market-Competitive Tiers: Three tiers - Basic , Growth , and Enterprise - were created to cater to different customer scales, from initial evaluation to high-volume production.
- Defining Transparent Usage Units: Billing was tied to logical, "customer-understandable" units such as API requests and data stored, rather than technical infrastructure metrics, to ensure transparency and reduce billing disputes.
- Streamlining Procurement: By integrating with the AWS Marketplace, the product leverages AWS-native procurement, allowing customers to subscribe easily and pay through their existing AWS invoice.
The integration component built by Ankercloud during the process helped Perception
Grid in the following ways:
a) Establish an event-Driven AWS Marketplace Integration & Secure Webhook
Orchestration
b) Automated Tenant Lifecycle Management & Dynamic Entitlement Enforcement
- Outsourcing Financial Operations: The vendor utilizes AWS to handle the heavy lifting of invoicing, payment collections, and disbursements, removing the need for a separate billing relationship with each customer.
- Ensuring Auditability and Fairness: The business committed to maintaining internal usage records to support accurate billing, resolve any potential disputes, and provide a clear audit trail for customers.
- Prioritizing Security as a Business Value: The listing emphasizes that customer data is logically isolated and never exposed to the marketplace systems, maintaining a high standard of security and trust.
