Blog — engineering first-hand

Predictive autoscaling, AI agents in production, cloud architecture. Notes from systems handling millions of requests a day — no marketing, with the numbers.

Serverless direct-to-S3 uploads — JWT authorization, presigned URLs, and access-control scoping

A file-upload API where bytes never touch the compute tier: a JWT authorizer plus S3 presigned URLs. The authentication-vs-authorization trap (a real BOLA fix), per-object state on S3 tags instead of a database, and multipart for large files.

Browser-based Android Automotive emulation — WebRTC media and on-demand provisioning

Delivering a car HMI (AAOS/Cuttlefish) to a browser tab with bidirectional audio. Why a "secure" SSM tunnel killed the sound (WebRTC media is UDP), fail-closed network access, and self-service provisioning of bare metal.

An AI agent on the operations team — what production taught us

Adopting an AI agent in operations is 20% a model problem and 80% a trust problem. Read-only by default, approval gates, a full audit trail — and only then intelligence. Notes straight from production.

Autoscaling that anticipates traffic — from 5.2 to 1.4 minutes of reaction time

Reactive autoscaling is always late, because CPU is a lagging indicator. How database connections and IOPS predict load 5–15 minutes ahead — and how that produced a 73% faster reaction.

New posts straight to your reader

The blog has an RSS feed — no algorithms, no spam.

Subscribe via RSS