// blog
Notes & writing.
Thoughts on engineering, data, AI - and the occasional dispatch from altitude.
// latest
18 Jul 2026/12 min read
How I Cut OCR API Costs by 80% Without Touching the Downstream
An OCR request costs you twice — once to the vendor for the call, once to your own machines for the work of producing it. I went after both: page merging to collapse the API calls, and a streamed, bounded pipeline to flatten RAM, CPU, and latency underneath them. Here's exactly how the combined cost reduction works.
EngineeringOCRPythonPerformanceCost Optimization
Read post
// more posts
