PichaFlow
Edge-Native Media Transformation & Delivery Engine
10ms - 50ms Edge Execution Time
0% Worker Out-Of-Memory Crashes
85% Bandwidth Cost Savings via WebP/AVIF
Project Overview
PichaFlow is a next-generation Digital Asset Management (DAM) SaaS built to serve media pipelines for global e-commerce systems, providing dynamic edge resizing, formatting, and secure delivery.
The Business Challenge
Executing dynamic, WebAssembly-based image transformations (like resizing, watermarking, and chroma-keying) on massive user-uploaded DSLR master files regularly caused Out-of-Memory (OOM) crashes in strict 128MB RAM Cloudflare Worker isolates. Additionally, invalidating global CDN caches for deleted image families (all resized permutations) was slow and expensive.
Our Solution & Engineering
We restructured the backend into decoupled micro-workers: one for asset ingestion (upload) and another for transformation (fetch). We implemented client-side pre-optimization using the HTML5 Canvas API in the PichaFlow SDK to downscale and transcode images to WebP prior to upload. For background removal, we implemented a custom Vlahos-lite ratio-based chroma-keying algorithm to bypass heavy math calculations under the 10ms edge CPU cap. Finally, we engineered a tombstone invalidation system using Cloudflare KV namespace caches to return instant 404s for deleted assets globally.
Key Integrations
Client-Side Pre-Optimization
Pre-scales massive images in the browser and dynamically loads dynamic modules like heic2any for HEIC assets.
Edge-Native Chroma Keying
Performs instant background removal at the edge using optimized Vlahos-lite ratios and Manhattan distance calculations.
Tombstone CDN Purges
Ensures compliance with GDPR/CCPA data deletion rules by creating edge tombstones to immediately drop CDN delivery.
Want similar results for your product?
Let’s discuss your technical bottlenecks and build a scalable solution tailored to your operational metrics.
Work With Us