Clip2Catalog

Challenges we ran into Optimizing Video Processing Latency: Processing raw video files directly inside serverless functions initially caused timeouts due to payload sizes. We solved this by implementing smart chunking and extracting keyframes asynchronously before sending lightweight payloads to Amazon Bedrock. Ensuring Marketplace Compliance: AI-generated text sometimes lacked strict compliance with Amazon's rigorous restricted-item and listing guidelines. We overcame this by engineering strict system prompts and multi-stage validation checks within our Bedrock pipeline to flag policy violations before publishing. Accomplishments that we're proud of Achieving an end-to-end pipeline speed where a raw 30-second social media review is parsed, verified, and converted into a complete, Amazon-compliant catalog listing in under 5 seconds. Building a clean, production-ready serverless architecture entirely anchored in native AWS services (Bedrock, Lambda, DynamoDB, S3, Amplify). What we learned Mastering serverless orchestration with AWS Lambda and API Gateway for handling asynchronous media streams. Fine-tuning multimodal prompts in Amazon Bedrock to extract high-precision JSON catalog schemas from unstructured social media video content without hallucinations. The critical importance of balancing automated AI generation with strict compliance validation rules in e-commerce tech. What's next for Clip2Catalog Multi-Marketplace Expansion: Expanding our ingestion engine to automatically format and push listings to other major e-commerce platforms beyond Amazon. Direct Seller API Integration: Integrating directly with the Amazon Selling Partner API (SP-API) so sellers can push generated listings directly to their live seller central accounts with a single click.

Built With

• amazon-bedrock
• amazon-dynamodb
• amazon-web-services
• aws-amplify