Specialized agents automating support triage, analytics, and fraud review
Fully deployed platform Rockwallet can operate and extend independently
Disruption to existing systems, CI/CD, or infrastructure model
Rockwallet
Waterloo, Canada
Cryptocurrency platform
Amazon Bedrock, Amazon OpenSearch, Amazon Redshift, Amazon Athena, AWS ECS, Amazon S3, AWS Strands Agents, Langfuse, Terraform
Rockwallet is a cryptocurrency platform focused on enabling secure and compliant digital asset transactions for individuals around the world. As the company’s data ecosystem matured, internal support workflows remained largely manual and difficult to scale, and the analytical potential of their infrastructure went underutilized. Avahi partnered with Rockwallet to design and deliver a modular AI assistant framework on AWS, deploying specialized agents that automate L2 support ticket triage, internal knowledge retrieval, transaction data analysis, and fraud review. The platform is production-ready, extensible, and built to grow alongside Rockwallet’s AI roadmap without disrupting existing systems.
Rockwallet is a cryptocurrency platform headquartered in Waterloo, Canada, that enables individuals to buy, sell, and swap digital assets securely and in compliance with financial regulations. With a strong focus on data integrity, regulatory compliance, and operational efficiency, Rockwallet has been building out its analytics and AI capabilities to better support both its internal teams and its growing customer base.
Rockwallet’s internal support operation relied heavily on manual effort. L2 support tickets required individual attention to route, research, and resolve, and the institutional knowledge needed to handle them quickly was scattered across historical ticket archives, internal documentation, and code repositories. As transaction volume grew, this approach was difficult to scale without proportionally growing the team.
At the same time, Rockwallet had invested in building a rich data ecosystem spanning structured databases, a data lake, and a support platform, yet extracting actionable insights from that infrastructure required engineering time the team could not always spare. Natural-language querying, rapid document lookup, and automated signal detection for fraud review were all within reach technically but remained out of reach operationally. Without a more automated approach, the gap between the team’s data assets and their ability to act on them would only widen.
Rockwallet’s existing infrastructure was built on AWS, making the cloud the natural foundation for any AI expansion. AWS Bedrock provided access to foundation models without the operational overhead of managing model infrastructure, while Amazon OpenSearch, Amazon Redshift, and Amazon Athena integrated directly into Rockwallet’s existing environment. The AWS Strands framework for model-first agent orchestration aligned with Rockwallet’s preference for tools that follow clean, configurable patterns and fit natively into their CI/CD practices.
As a premier-tier AWS partner with deep expertise in cloud-native AI and data engineering, Avahi brought the combination of architectural depth and delivery capability that Rockwallet needed to move from vision to working agents within a defined timeline. Avahi’s approach centered on augmenting Rockwallet’s existing systems rather than replacing them, designing a modular platform that the Rockwallet team could own, operate, and extend independently after handoff.
Avahi’s experience with the Strands orchestration framework and AWS Bedrock, combined with a structured engagement covering design, implementation, evaluation, and full documentation, gave Rockwallet confidence that the delivered platform would be production-ready and maintainable long-term.
Avahi designed and delivered a layered AI agent platform built on AWS Bedrock and the Strands model-first orchestration framework. At the center of the architecture is an Operator layer that detects the intent behind incoming queries and routes them to the appropriate specialized agent, keeping the system modular and independently extensible.
Three specialized agents were deployed to address Rockwallet’s highest-priority operational needs. A knowledge and support retrieval agent uses retrieval-augmented generation (RAG) to query indexed historical ticket data and internal documentation stored in Amazon OpenSearch, enabling fast, grounded answers to L2 support queries without manual research. A transaction analytics agent connects to Rockwallet’s existing Redshift and Athena environments via the Redshift MCP server, allowing teams to run natural-language queries against their data lake. A fraud review agent surfaces relevant signals from structured data sources to support internal review workflows.
To make the platform production-ready, Langfuse was configured for trace logging, score inspection, and agent evaluation, giving Rockwallet’s team visibility into agent behavior and a feedback loop for continuous improvement. The entire stack was containerized and deployed on AWS ECS using Terraform, fitting into Rockwallet’s existing CI/CD pipeline without requiring changes to their infrastructure model.
Agent orchestration logic, permissions, and tool configurations are managed through YAML-based configuration files following the native Strands pattern, so Rockwallet engineers can add, adjust, or replace agents without modifying core framework code. A HubSpot ticket ingestion pipeline was also delivered, enabling the support agent to ingest and query live ticket streams through a webhook-based integration. Historical ticket data was loaded into Amazon OpenSearch to ground the RAG agent in Rockwallet’s actual support history.
The engagement delivered a production-grade AI platform that Rockwallet can operate and extend independently. Internal support queries that previously required manual research can now be routed to specialized agents grounded in indexed ticket history and internal documentation, reducing the time teams spend on repetitive resolution tasks. The transaction analytics agent opens direct, natural-language access to Rockwallet’s data lake for internal stakeholders, without requiring engineering intervention for each query. The modular architecture and YAML-based configuration model ensure that adding new agents or data sources does not require rebuilding the core platform, positioning Rockwallet to scale their internal AI capabilities alongside business growth.
Rockwallet
Waterloo, Canada
Cryptocurrency platform
Amazon Bedrock, Amazon OpenSearch, Amazon Redshift, Amazon Athena, AWS ECS, Amazon S3, AWS Strands Agents, Langfuse, Terraform
Rockwallet is a cryptocurrency platform focused on enabling secure and compliant digital asset transactions for individuals around the world. As the company’s data ecosystem matured, internal support workflows remained largely manual and difficult to scale, and the analytical potential of their infrastructure went underutilized. Avahi partnered with Rockwallet to design and deliver a modular AI assistant framework on AWS, deploying specialized agents that automate L2 support ticket triage, internal knowledge retrieval, transaction data analysis, and fraud review. The platform is production-ready, extensible, and built to grow alongside Rockwallet’s AI roadmap without disrupting existing systems.
Rockwallet is a cryptocurrency platform headquartered in Waterloo, Canada, that enables individuals to buy, sell, and swap digital assets securely and in compliance with financial regulations. With a strong focus on data integrity, regulatory compliance, and operational efficiency, Rockwallet has been building out its analytics and AI capabilities to better support both its internal teams and its growing customer base.
Rockwallet’s internal support operation relied heavily on manual effort. L2 support tickets required individual attention to route, research, and resolve, and the institutional knowledge needed to handle them quickly was scattered across historical ticket archives, internal documentation, and code repositories. As transaction volume grew, this approach was difficult to scale without proportionally growing the team.
At the same time, Rockwallet had invested in building a rich data ecosystem spanning structured databases, a data lake, and a support platform, yet extracting actionable insights from that infrastructure required engineering time the team could not always spare. Natural-language querying, rapid document lookup, and automated signal detection for fraud review were all within reach technically but remained out of reach operationally. Without a more automated approach, the gap between the team’s data assets and their ability to act on them would only widen.
Rockwallet’s existing infrastructure was built on AWS, making the cloud the natural foundation for any AI expansion. AWS Bedrock provided access to foundation models without the operational overhead of managing model infrastructure, while Amazon OpenSearch, Amazon Redshift, and Amazon Athena integrated directly into Rockwallet’s existing environment. The AWS Strands framework for model-first agent orchestration aligned with Rockwallet’s preference for tools that follow clean, configurable patterns and fit natively into their CI/CD practices.
As a premier-tier AWS partner with deep expertise in cloud-native AI and data engineering, Avahi brought the combination of architectural depth and delivery capability that Rockwallet needed to move from vision to working agents within a defined timeline. Avahi’s approach centered on augmenting Rockwallet’s existing systems rather than replacing them, designing a modular platform that the Rockwallet team could own, operate, and extend independently after handoff.
Avahi’s experience with the Strands orchestration framework and AWS Bedrock, combined with a structured engagement covering design, implementation, evaluation, and full documentation, gave Rockwallet confidence that the delivered platform would be production-ready and maintainable long-term.
Avahi designed and delivered a layered AI agent platform built on AWS Bedrock and the Strands model-first orchestration framework. At the center of the architecture is an Operator layer that detects the intent behind incoming queries and routes them to the appropriate specialized agent, keeping the system modular and independently extensible.
Three specialized agents were deployed to address Rockwallet’s highest-priority operational needs. A knowledge and support retrieval agent uses retrieval-augmented generation (RAG) to query indexed historical ticket data and internal documentation stored in Amazon OpenSearch, enabling fast, grounded answers to L2 support queries without manual research. A transaction analytics agent connects to Rockwallet’s existing Redshift and Athena environments via the Redshift MCP server, allowing teams to run natural-language queries against their data lake. A fraud review agent surfaces relevant signals from structured data sources to support internal review workflows.
To make the platform production-ready, Langfuse was configured for trace logging, score inspection, and agent evaluation, giving Rockwallet’s team visibility into agent behavior and a feedback loop for continuous improvement. The entire stack was containerized and deployed on AWS ECS using Terraform, fitting into Rockwallet’s existing CI/CD pipeline without requiring changes to their infrastructure model.
Agent orchestration logic, permissions, and tool configurations are managed through YAML-based configuration files following the native Strands pattern, so Rockwallet engineers can add, adjust, or replace agents without modifying core framework code. A HubSpot ticket ingestion pipeline was also delivered, enabling the support agent to ingest and query live ticket streams through a webhook-based integration. Historical ticket data was loaded into Amazon OpenSearch to ground the RAG agent in Rockwallet’s actual support history.
The engagement delivered a production-grade AI platform that Rockwallet can operate and extend independently. Internal support queries that previously required manual research can now be routed to specialized agents grounded in indexed ticket history and internal documentation, reducing the time teams spend on repetitive resolution tasks. The transaction analytics agent opens direct, natural-language access to Rockwallet’s data lake for internal stakeholders, without requiring engineering intervention for each query. The modular architecture and YAML-based configuration model ensure that adding new agents or data sources does not require rebuilding the core platform, positioning Rockwallet to scale their internal AI capabilities alongside business growth.
Let’s explore your high-impact AI opportunities together in a complimentary session
