Three cloud vendors consolidated into one AWS footprint
Daily jobs moved to durable, replayable SQS queuing
Availability targets established, from no prior SLA commitments
Profound
New York
AI Brand Analytics Software
Amazon ECS, Amazon EC2 Spot Instances, Amazon SQS, Amazon EventBridge, Amazon MSK, Amazon RDS for PostgreSQL, Amazon MemoryDB for Redis, Amazon DynamoDB, Amazon S3, Amazon ECR, AWS Secrets Manager, Amazon CloudWatch, AWS CloudTrail, AWS Compute Savings Plans, Terraform, GitHub Actions
Profound, a rapidly growing AI brand intelligence software company, had built its core data ingestion and analytics platform across three separate cloud vendors: a large-scale container compute platform for LLM-driven data collection, a platform-as-a-service provider for real-time analytics, and Google Cloud Platform for internal AI workflow tooling. While this multi-platform model enabled rapid early growth, it created mounting reliability risks, accelerating costs, and no path to operational maturity. Avahi conducted a comprehensive migration assessment across all three environments, designed and validated a consolidated AWS target architecture, and executed a wave-based migration of the full platform — moving workloads processing 30 million jobs per day with 2.5 TB of daily network egress to Amazon ECS on EC2 Spot Instances, Amazon SQS, and Amazon MSK, replacing best-effort queuing with durable, replayable messaging and consolidating three disconnected environments into a single Terraform-managed AWS footprint.
Profound is an AI brand intelligence company that equips global brands with visibility into how their narratives appear inside AI-powered search experiences. The company’s Conversation Explorer, AI Leaderboard, and Answer Engine Insights products give marketing and brand teams a real-time window into how AI systems — including major search engines and generative AI platforms — describe, reference, and position their brands. Profound’s platform continuously ingests and analyzes AI-generated content at scale, making infrastructure reliability and data durability central to its product promise.
Profound’s infrastructure was distributed across three separate cloud environments, each with distinct operational models and limitations. The primary compute environment handled LLM perception workloads — over 4,000 concurrent containers processing 30 million jobs per day across 10 AI search collector types and multiple LLM processor pipelines — but offered no message acknowledgement or replay capability. Orphaned jobs and the inability to reprocess data were persistent reliability risks that directly affected the analytics products customers depended on.
Infrastructure costs were accelerating without a clear ceiling. Monthly spend on the primary compute platform grew 427% in five months, from approximately $17,700 in June 2025 to a projected $93,000 by November, driven by container compute, egress, and GPU usage, while the overall multi-platform run rate reached $101,500 per month. Without intervention, that growth trajectory would have exceeded a comparable AWS architecture before year-end while retaining all the reliability and observability gaps.
The operational posture compounded the risk. All three environments were managed through vendor consoles with no infrastructure-as-code, no environment parity between development, staging, and production, no disaster recovery capability, and no incident runbooks. Profound had no formal SLAs with its enterprise customers, but commercial growth was making that posture increasingly untenable.
AWS offered the combination of durable messaging primitives, cost-effective container compute, and a comprehensive observability stack that Profound’s platform required. Amazon SQS with dead-letter queues and Amazon EventBridge replaced the vendor queuing model with durable, replayable message delivery — directly eliminating the primary reliability failure mode. Amazon ECS on EC2 Spot Instances replicated the auto-scaling, long-running poller pattern Profound’s collectors depended on, while EC2’s public IP addressing model eliminated the NAT gateway costs that made alternative container hosting economically infeasible at 2.5 TB of daily egress. Amazon MSK provided managed throughput replacement for the existing Kafka layer. Across the data tier, Amazon RDS for PostgreSQL with Multi-AZ deployment, Amazon MemoryDB for Redis, and Amazon DynamoDB for idempotency tracking replaced vendor-managed databases with services that carried production-grade SLA commitments. Terraform coverage across all AWS resources established the infrastructure-as-code foundation that was entirely absent before migration.
As an AWS Premier Tier Services Partner, Avahi brought the technical depth in cloud migration architecture and the AWS Migration Acceleration Program (MAP) expertise needed to assess and execute a migration of this operational complexity. Profound’s platform combined high-throughput containerized compute, stream processing, managed databases, and AI workflow tooling across three separate vendors — a scope requiring architecture validation grounded in actual billing data, workload measurements, and stakeholder alignment before any migration work began. Avahi’s MAP Assessment framework gave Profound a structured, evidence-based path from multi-platform fragmentation to a validated AWS architecture and execution plan, with MAP funding support reducing the cost of the engagement.
Avahi began with a thorough MAP Assessment of Profound’s three-platform environment, combining documentation review, billing analysis spanning ten months of production data, stakeholder working sessions, and architecture review with AWS Solution Architecture support. The assessment catalogued all compute, messaging, data, and networking resources across each platform, quantified cost drivers and trajectory, and produced a full current-to-AWS service mapping validated against AWS Pricing Calculator outputs.
The target architecture consolidated Profound’s entire platform onto Amazon ECS running EC2 Spot Instances across a three-account AWS landing zone covering production, staging, and development environments — the first time Profound had dedicated non-production environments. Amazon SQS and Amazon EventBridge replaced the existing queuing system, introducing message acknowledgement, dead-letter queues, and full replay capability for the first time. The Kafka layer was migrated to Amazon MSK for the analytics ingestion pipeline. Amazon RDS for PostgreSQL (Multi-AZ), Amazon MemoryDB for Redis with automatic failover, and Amazon DynamoDB provided the data tier. Terraform modules covering VPC, compute, data services, messaging, and observability were authored as reusable building blocks across all three environments. GitHub Actions pipelines automated infrastructure promotion: pull requests triggered Terraform plans without applying, merges to main deployed to development automatically, and workflow dispatch with approval gates promoted to staging and production.
Migration executed in three waves aligned to business priority. Wave 1 moved the highest-criticality collector and processor workloads to AWS by end of November 2025, eliminating the data-loss risk that represented the most urgent business exposure. Wave 2 migrated the analytics services — FastAPI, Kafka consumers, and Redis — onto the same ECS foundation. Wave 3 completed the engagement by migrating the GCP-hosted internal AI workflow services. Throughout all waves, dual-run operations were maintained and rollback paths were documented before each cutover, with CloudFlare CNAME changes providing controlled, per-domain traffic switching. No historical data migration was required, simplifying the cutover timeline and reducing operational risk across all waves.
Profound consolidated its AI brand intelligence platform from three separate cloud vendors onto a single AWS infrastructure footprint, replacing manual console-driven operations with a fully infrastructure-as-code deployment model. The migration delivered the operational reliability foundation Profound’s commercial growth required.
Durable Amazon SQS queuing with replay and dead-letter queues replaced the queuing model that had produced orphaned jobs and prevented historical data reprocessing, directly eliminating the primary reliability incident driver. The AWS multi-AZ ECS architecture established 99.9% availability targets for collection and processing workloads and 99.95% for the data layer — from a prior baseline of no SLA commitments on any workload. The EC2 public IP addressing strategy preserved 2.5 TB of daily production egress while eliminating NAT gateway fees, and the 90% Spot / 10% Compute Savings Plans compute mix established a cost model with clear optimization levers that the previous vendor environments did not provide.
The engagement established Profound’s first dedicated development and staging environments, replacing a production-only posture with isolated, identically structured environments — a prerequisite for safe deployments at enterprise scale. Infrastructure promotion from development to staging to production became a pull-request and approval-gate workflow, replacing manual click-operations with full audit trails.
Profound
New York
AI Brand Analytics Software
Amazon ECS, Amazon EC2 Spot Instances, Amazon SQS, Amazon EventBridge, Amazon MSK, Amazon RDS for PostgreSQL, Amazon MemoryDB for Redis, Amazon DynamoDB, Amazon S3, Amazon ECR, AWS Secrets Manager, Amazon CloudWatch, AWS CloudTrail, AWS Compute Savings Plans, Terraform, GitHub Actions
Profound, a rapidly growing AI brand intelligence software company, had built its core data ingestion and analytics platform across three separate cloud vendors: a large-scale container compute platform for LLM-driven data collection, a platform-as-a-service provider for real-time analytics, and Google Cloud Platform for internal AI workflow tooling. While this multi-platform model enabled rapid early growth, it created mounting reliability risks, accelerating costs, and no path to operational maturity. Avahi conducted a comprehensive migration assessment across all three environments, designed and validated a consolidated AWS target architecture, and executed a wave-based migration of the full platform — moving workloads processing 30 million jobs per day with 2.5 TB of daily network egress to Amazon ECS on EC2 Spot Instances, Amazon SQS, and Amazon MSK, replacing best-effort queuing with durable, replayable messaging and consolidating three disconnected environments into a single Terraform-managed AWS footprint.
Profound is an AI brand intelligence company that equips global brands with visibility into how their narratives appear inside AI-powered search experiences. The company’s Conversation Explorer, AI Leaderboard, and Answer Engine Insights products give marketing and brand teams a real-time window into how AI systems — including major search engines and generative AI platforms — describe, reference, and position their brands. Profound’s platform continuously ingests and analyzes AI-generated content at scale, making infrastructure reliability and data durability central to its product promise.
Profound’s infrastructure was distributed across three separate cloud environments, each with distinct operational models and limitations. The primary compute environment handled LLM perception workloads — over 4,000 concurrent containers processing 30 million jobs per day across 10 AI search collector types and multiple LLM processor pipelines — but offered no message acknowledgement or replay capability. Orphaned jobs and the inability to reprocess data were persistent reliability risks that directly affected the analytics products customers depended on.
Infrastructure costs were accelerating without a clear ceiling. Monthly spend on the primary compute platform grew 427% in five months, from approximately $17,700 in June 2025 to a projected $93,000 by November, driven by container compute, egress, and GPU usage, while the overall multi-platform run rate reached $101,500 per month. Without intervention, that growth trajectory would have exceeded a comparable AWS architecture before year-end while retaining all the reliability and observability gaps.
The operational posture compounded the risk. All three environments were managed through vendor consoles with no infrastructure-as-code, no environment parity between development, staging, and production, no disaster recovery capability, and no incident runbooks. Profound had no formal SLAs with its enterprise customers, but commercial growth was making that posture increasingly untenable.
AWS offered the combination of durable messaging primitives, cost-effective container compute, and a comprehensive observability stack that Profound’s platform required. Amazon SQS with dead-letter queues and Amazon EventBridge replaced the vendor queuing model with durable, replayable message delivery — directly eliminating the primary reliability failure mode. Amazon ECS on EC2 Spot Instances replicated the auto-scaling, long-running poller pattern Profound’s collectors depended on, while EC2’s public IP addressing model eliminated the NAT gateway costs that made alternative container hosting economically infeasible at 2.5 TB of daily egress. Amazon MSK provided managed throughput replacement for the existing Kafka layer. Across the data tier, Amazon RDS for PostgreSQL with Multi-AZ deployment, Amazon MemoryDB for Redis, and Amazon DynamoDB for idempotency tracking replaced vendor-managed databases with services that carried production-grade SLA commitments. Terraform coverage across all AWS resources established the infrastructure-as-code foundation that was entirely absent before migration.
As an AWS Premier Tier Services Partner, Avahi brought the technical depth in cloud migration architecture and the AWS Migration Acceleration Program (MAP) expertise needed to assess and execute a migration of this operational complexity. Profound’s platform combined high-throughput containerized compute, stream processing, managed databases, and AI workflow tooling across three separate vendors — a scope requiring architecture validation grounded in actual billing data, workload measurements, and stakeholder alignment before any migration work began. Avahi’s MAP Assessment framework gave Profound a structured, evidence-based path from multi-platform fragmentation to a validated AWS architecture and execution plan, with MAP funding support reducing the cost of the engagement.
Avahi began with a thorough MAP Assessment of Profound’s three-platform environment, combining documentation review, billing analysis spanning ten months of production data, stakeholder working sessions, and architecture review with AWS Solution Architecture support. The assessment catalogued all compute, messaging, data, and networking resources across each platform, quantified cost drivers and trajectory, and produced a full current-to-AWS service mapping validated against AWS Pricing Calculator outputs.
The target architecture consolidated Profound’s entire platform onto Amazon ECS running EC2 Spot Instances across a three-account AWS landing zone covering production, staging, and development environments — the first time Profound had dedicated non-production environments. Amazon SQS and Amazon EventBridge replaced the existing queuing system, introducing message acknowledgement, dead-letter queues, and full replay capability for the first time. The Kafka layer was migrated to Amazon MSK for the analytics ingestion pipeline. Amazon RDS for PostgreSQL (Multi-AZ), Amazon MemoryDB for Redis with automatic failover, and Amazon DynamoDB provided the data tier. Terraform modules covering VPC, compute, data services, messaging, and observability were authored as reusable building blocks across all three environments. GitHub Actions pipelines automated infrastructure promotion: pull requests triggered Terraform plans without applying, merges to main deployed to development automatically, and workflow dispatch with approval gates promoted to staging and production.
Migration executed in three waves aligned to business priority. Wave 1 moved the highest-criticality collector and processor workloads to AWS by end of November 2025, eliminating the data-loss risk that represented the most urgent business exposure. Wave 2 migrated the analytics services — FastAPI, Kafka consumers, and Redis — onto the same ECS foundation. Wave 3 completed the engagement by migrating the GCP-hosted internal AI workflow services. Throughout all waves, dual-run operations were maintained and rollback paths were documented before each cutover, with CloudFlare CNAME changes providing controlled, per-domain traffic switching. No historical data migration was required, simplifying the cutover timeline and reducing operational risk across all waves.
Profound consolidated its AI brand intelligence platform from three separate cloud vendors onto a single AWS infrastructure footprint, replacing manual console-driven operations with a fully infrastructure-as-code deployment model. The migration delivered the operational reliability foundation Profound’s commercial growth required.
Durable Amazon SQS queuing with replay and dead-letter queues replaced the queuing model that had produced orphaned jobs and prevented historical data reprocessing, directly eliminating the primary reliability incident driver. The AWS multi-AZ ECS architecture established 99.9% availability targets for collection and processing workloads and 99.95% for the data layer — from a prior baseline of no SLA commitments on any workload. The EC2 public IP addressing strategy preserved 2.5 TB of daily production egress while eliminating NAT gateway fees, and the 90% Spot / 10% Compute Savings Plans compute mix established a cost model with clear optimization levers that the previous vendor environments did not provide.
The engagement established Profound’s first dedicated development and staging environments, replacing a production-only posture with isolated, identically structured environments — a prerequisite for safe deployments at enterprise scale. Infrastructure promotion from development to staging to production became a pull-request and approval-gate workflow, replacing manual click-operations with full audit trails.
Let’s explore your high-impact AI opportunities together in a complimentary session
