Data loss on the production database cutover; the legacy environment was fully decommissioned afterward
Production security groups audited to least-privilege, closing one untracked access path along the way
Application platform containerized and moved onto a production-ready Amazon EKS foundation
Boon AI
California, United States
Construction Technology, Computer Vision, AI/ML
Amazon EKS, Amazon RDS, Amazon ElastiCache for Valkey, Amazon ECR, AWS IAM, AWS CloudTrail, Amazon GuardDuty
Boon AI’s platform automates construction takeoff and project estimation, turning a process that used to take days into minutes for general contractors and subcontractors. As the platform scaled, its infrastructure had outgrown the mix of services it launched on. Boon AI partnered with Avahi, an AWS Premier Tier Services Partner, to rebuild its application and data platform on AWS — a hardened account foundation, a production-ready Kubernetes environment, and a fully migrated database layer, delivered with zero disruption to the service contractors rely on every day.
Boon AI is a construction technology company. Its computer-vision and LLM-powered platform automates takeoff and project estimation for general contractors and subcontractors across the United States.
Boon AI’s platform had grown quickly on a mix of managed services, and its operational foundation hadn’t kept pace: no centralized audit trail, no account-level guardrails, and a database layer still running on its original provider. The team needed a hardened, production-grade AWS foundation without slowing down the product team building on top of it.
AWS gave Boon AI a single, unified control plane for application and data workloads in one place, along with the account-level guardrails — CloudTrail, GuardDuty, least-privilege IAM — a fast-moving platform needs as it scales toward larger customers.
Avahi had already run Boon AI’s MAP Assess phase, so the migration plan was built on its real production environment and there was no second discovery cycle. The plan also protected the product. No production change could happen without the engineering lead’s approval, and the database cutover carried a zero-data-loss requirement backed by a 48-hour parallel run. Avahi also covered Boon AI’s GPU inference and LLM layers alongside the application and data migration. As an AWS Premier Tier Services Partner, it secured AWS MAP funding for the engagement.
Avahi rebuilt Boon AI’s AWS foundation end to end. On governance and network, that meant a redesigned VPC across four availability zones, a full audit of every production security group, CloudTrail and GuardDuty brought under infrastructure-as-code, and MAP-compliant resource tagging. On the application side, Avahi containerized every Boon service, built a production image supply chain on Amazon ECR with GitHub Actions and OIDC, and stood up Amazon EKS with Karpenter for dynamic, cost-aware scaling. On the data side, Avahi migrated Boon AI’s production PostgreSQL database to Amazon RDS — delivered on a newer Graviton-based instance for better price-performance — and its cache layer to Amazon ElastiCache for Valkey, executing the cutover under a zero-data-loss mandate with a tested rollback runbook.
Avahi also ran a hands-on technical evaluation of Boon AI’s GPU inference workloads against AWS Inferentia2, rather than defaulting to a lift-and-shift. The evaluation found that Boon AI’s computer-vision models rely on dynamic input shapes and custom CUDA kernels that don’t compile to Inferentia2’s architecture, and recommended a cost-optimization path instead: reserved-instance commitments, model quantization, and workload right-sizing.
Hardened AWS account foundation: CloudTrail, GuardDuty, least-privilege IAM, protected Terraform state, MAP-compliant tagging
Redesigned VPC and network security, with every production security group audited
Production-ready Amazon EKS environment with Karpenter autoscaling and a complete Amazon ECR / GitHub Actions image pipeline
PostgreSQL and cache migration to Amazon RDS and Amazon ElastiCache for Valkey, cut over with zero data loss
CloudWatch and Grafana observability across infrastructure and application layers
A GPU cost-optimization strategy grounded in hands-on model-compatibility testing
Final as-built architecture documentation and operations runbooks handed to Boon AI’s team
With the migration complete, Boon AI runs its application and data platform on a single, auditable AWS foundation instead of a patchwork of managed services. The team has account-level guardrails and a full security-group audit behind it, a production Kubernetes environment built to scale with demand, and a database layer it fully owns on AWS. The GPU cost evaluation gave Boon AI a validated, data-backed path to lower inference costs without risking the accuracy of the computer-vision models its platform depends on.
Zero data loss on the production database cutover
16 production security groups audited; 1 untracked access path closed
100% of application services containerized and running on Amazon EKS
Boon AI
California, United States
Construction Technology, Computer Vision, AI/ML
Amazon EKS, Amazon RDS, Amazon ElastiCache for Valkey, Amazon ECR, AWS IAM, AWS CloudTrail, Amazon GuardDuty
Boon AI’s platform automates construction takeoff and project estimation, turning a process that used to take days into minutes for general contractors and subcontractors. As the platform scaled, its infrastructure had outgrown the mix of services it launched on. Boon AI partnered with Avahi, an AWS Premier Tier Services Partner, to rebuild its application and data platform on AWS — a hardened account foundation, a production-ready Kubernetes environment, and a fully migrated database layer, delivered with zero disruption to the service contractors rely on every day.
Boon AI is a construction technology company. Its computer-vision and LLM-powered platform automates takeoff and project estimation for general contractors and subcontractors across the United States.
Boon AI’s platform had grown quickly on a mix of managed services, and its operational foundation hadn’t kept pace: no centralized audit trail, no account-level guardrails, and a database layer still running on its original provider. The team needed a hardened, production-grade AWS foundation without slowing down the product team building on top of it.
AWS gave Boon AI a single, unified control plane for application and data workloads in one place, along with the account-level guardrails — CloudTrail, GuardDuty, least-privilege IAM — a fast-moving platform needs as it scales toward larger customers.
Avahi had already run Boon AI’s MAP Assess phase, so the migration plan was built on its real production environment and there was no second discovery cycle. The plan also protected the product. No production change could happen without the engineering lead’s approval, and the database cutover carried a zero-data-loss requirement backed by a 48-hour parallel run. Avahi also covered Boon AI’s GPU inference and LLM layers alongside the application and data migration. As an AWS Premier Tier Services Partner, it secured AWS MAP funding for the engagement.
Avahi rebuilt Boon AI’s AWS foundation end to end. On governance and network, that meant a redesigned VPC across four availability zones, a full audit of every production security group, CloudTrail and GuardDuty brought under infrastructure-as-code, and MAP-compliant resource tagging. On the application side, Avahi containerized every Boon service, built a production image supply chain on Amazon ECR with GitHub Actions and OIDC, and stood up Amazon EKS with Karpenter for dynamic, cost-aware scaling. On the data side, Avahi migrated Boon AI’s production PostgreSQL database to Amazon RDS — delivered on a newer Graviton-based instance for better price-performance — and its cache layer to Amazon ElastiCache for Valkey, executing the cutover under a zero-data-loss mandate with a tested rollback runbook.
Avahi also ran a hands-on technical evaluation of Boon AI’s GPU inference workloads against AWS Inferentia2, rather than defaulting to a lift-and-shift. The evaluation found that Boon AI’s computer-vision models rely on dynamic input shapes and custom CUDA kernels that don’t compile to Inferentia2’s architecture, and recommended a cost-optimization path instead: reserved-instance commitments, model quantization, and workload right-sizing.
Hardened AWS account foundation: CloudTrail, GuardDuty, least-privilege IAM, protected Terraform state, MAP-compliant tagging
Redesigned VPC and network security, with every production security group audited
Production-ready Amazon EKS environment with Karpenter autoscaling and a complete Amazon ECR / GitHub Actions image pipeline
PostgreSQL and cache migration to Amazon RDS and Amazon ElastiCache for Valkey, cut over with zero data loss
CloudWatch and Grafana observability across infrastructure and application layers
A GPU cost-optimization strategy grounded in hands-on model-compatibility testing
Final as-built architecture documentation and operations runbooks handed to Boon AI’s team
With the migration complete, Boon AI runs its application and data platform on a single, auditable AWS foundation instead of a patchwork of managed services. The team has account-level guardrails and a full security-group audit behind it, a production Kubernetes environment built to scale with demand, and a database layer it fully owns on AWS. The GPU cost evaluation gave Boon AI a validated, data-backed path to lower inference costs without risking the accuracy of the computer-vision models its platform depends on.
Zero data loss on the production database cutover
16 production security groups audited; 1 untracked access path closed
100% of application services containerized and running on Amazon EKS
Zero data loss on the production database cutover
16 production security groups audited; 1 untracked access path closed
100% of application services containerized and running on Amazon EKS
Let’s explore your high-impact AI opportunities together in a complimentary session
