VeUP
← All case studies
Production Engagement · Anonymized
VeUP designs a computer-vision inventory platform on AWS for an F&B ML startup
Food & Beverage (computer-vision / ML SaaS)
Amazon Kinesis Video StreamsAmazon RekognitionAmazon SageMakerAmazon Bedrock
Machine Learning · Computer Vision / ML Design

An Omani F&B distribution-prediction ML SaaS startup

AWS-validated reference — full details available to AWS Partner Validation or on request. (Customer name held on file with VeUP.)

For the customer, an Omani F&B distribution-prediction ML SaaS startup, VeUP authored a full real-time computer-vision design plus a costed AWS POC — Kinesis Video Streams to Lambda to Rekognition Custom Labels to DynamoDB, with SageMaker quality models and a Glue/Redshift/QuickSight tier. Scoped design.

The challenge

the customer, an Omani machine-learning SaaS startup, helps F&B companies predict product distribution at the customer level by connecting and processing their data sources. It needed an AWS reference architecture for automated inventory management, quality assessment, predictive analytics, and automated reordering via computer vision — plus a costed POC entry point — to scale from a small enterprise base.

The solution

VeUP authored a full computer-vision solution design and a costed AWS Pricing Calculator POC. The designed pipeline ingests camera feeds through Amazon Kinesis Video Streams to AWS Lambda to Amazon Rekognition Custom Labels (SKU/brand recognition) to Amazon DynamoDB inventory state; Amazon SageMaker runs damage/quality models; AWS Glue unifies product databases (GS1, Open Food Facts, USDA, distributor catalogs) into Amazon Redshift with Amazon QuickSight dashboards and Amazon Forecast demand modeling; AWS Step Functions orchestrate automated reordering; and an Amazon Bedrock layer provides conversational analytics over the data so users ask questions instead of reading charts.

Production outcomes

KPIResult
Production outcomesA phased AWS reference architecture (four phases over ~10-12 months, 2-client POC entry) and a costed AWS Pricing Calculator estimate (~$25,066/month for a 2-client POC across Kinesis Video Streams, Rekognition Image, S3, Lambda, DynamoDB) were delivered. Design targets stated (detection accuracy >95%, processing <2s, uptime 99.9%) are design goals, not measured production results — the pipeline is a scoped design, not a launched workload.
Engagement window2025-02-20 (costed POC); CV Services opportunity at Onboarding → 2025-05-19 (design/POC delivered); Resell + Quicksight & AI opportunities Closed-Lost
Cost / TCO postureA costed AWS Pricing Calculator POC (EU/Frankfurt) projected ~$25,066/month (~$300,792/12mo) for a 2-client POC: Kinesis Video Streams (8 devices), Rekognition Image (500K images/month), S3, Lambda, and DynamoDB on-demand. This is a pre-build pricing estimate, not realized spend.
Lessons & continuationKinesis Video Streams is the natural ingest entry point for a real-time CV inventory pipeline; Rekognition Custom Labels handles SKU/brand recognition without bespoke model training; a Bedrock conversational layer over the analytics tier is what turns dashboards into self-serve answers.

Architecture

A Well-Architected-annotated view of the previous state and the target-state AWS computer-vision design — Kinesis Video Streams ingestion through Rekognition, SageMaker, and the analytics tier — mapped to the six AWS Well-Architected pillars. Design / costed-POC reference, not a production deployment.

AWS Well-Architected annotated architecture diagram: previous state (manual, disconnected camera feeds, product catalogs, ERP, and reporting) versus the target-state AWS design — Amazon Kinesis Video Streams ingest through AWS Lambda and Amazon Rekognition Custom Labels to Amazon DynamoDB, Amazon SageMaker quality models, AWS Glue-unified product catalogs into Amazon Redshift and Amazon QuickSight with Amazon Forecast demand modeling, AWS Step Functions automated reordering, and an Amazon Bedrock conversational-analytics layer, with VPC security controls and Amazon CloudWatch observability — with Well-Architected pillar callouts. Costed AWS Pricing Calculator POC reference, not a production launch.
AWS services in production

Amazon Kinesis Video Streams · AWS Lambda · Amazon Rekognition Custom Labels · Amazon DynamoDB · Amazon SageMaker · AWS Glue · Amazon Redshift · Amazon QuickSight · Amazon Forecast · AWS Step Functions · Amazon Bedrock