RAG & Embedding
Document Processing
Pulse Labs: Multi-modal model comparison, embedding, and video processing optimization
"Working with the Tech 42 team on our video analysis POC was a great experience. They brought a very thorough and systematic engineering approach to evaluating AI models for video understanding, helping us compare Twelve Labs Pegasus against Amazon Nova models and AWS Bedrock Data Automation in a real-world environment. The solution they built processed our sample session recordings end-to-end—from video ingestion and automated description generation to embeddings, storage, and search testing—giving us concrete performance data and a clear pipeline we can build on. Their documentation, transparency, and communication throughout the engagement made the handoff easy, and we now feel confident in the model choices and architecture needed to take the next step."
Drew Raines
Head of Product
Project summary
Tech 42 built a proof-of-concept system on Amazon Bedrock that automates video analysis and enables conversational search. The system ingests video recordings, generates automated descriptions, creates embeddings, and makes content searchable and queryable by natural language. The architecture uses AWS Lambda for processing, Amazon S3 for storage, and Amazon SageMaker for model evaluation. Tech 42 compared TwelveLabs Pegasus 1.2 against Amazon Nova models and assessed Amazon Bedrock Data Automation for content transformation. The POC delivered end-to-end video processing, a working search interface, and complete technical documentation. Pulse Labs now has concrete performance data and a production-ready architecture to build on.
