Cost-optimization

Published on
September 13, 2026
Don't Enrich Rows. Enrich Hashes: LLM Nutrition on DynamoDB
Architecture DynamoDB LLM OpenRouter Cost-Optimization Python Food-Delivery Data-Pipelines Caching
Menu nutrition is a unique-hash problem. Deduplicate by brand and dish, cache LLM rows locally, fan out onto one DynamoDB item per store, and keep spend on the unique set — a full marketplace catalog for about four dollars.
Published on
August 16, 2026
Semantic Search Without a Second Database: In-Process ONNX and DynamoDB Vectors
AI LLM Vector-Search Semantic-Search Embeddings ONNX DynamoDB AWS ECS Spring-Boot Architecture Cost-Optimization Java Performance Mobile Delivery
How a food-delivery catalog API on ECS adds semantic store and dish search: in-process bge-small ONNX, vectors on the same DynamoDB items, SearchVectors for retrieval, and a fail-open query parser. No embed sidecar, no second vector store.
Published on
August 5, 2026
Measuring and Cutting ECS Cost and Latency Across a Microservice Fleet
AWS ECS Performance Spring-Boot k6 Cost-Optimization
Smoke tests prove correctness. A before/after k6 harness, Fargate rightsizing, extracted JARs with AOT cache, and virtual threads prove speed and cost.

Don't Enrich Rows. Enrich Hashes: LLM Nutrition on DynamoDB