Open to staff and principal roles

Kraig Jacobson

I build the internal tooling other engineers depend on. Developer platforms, Kubernetes, Go and TypeScript, plus the AI infrastructure underneath a product I run myself.

Kraig Jacobson

The thing I am proudest of

Named inventor · Patent pending · Filed March 2026

A week of setup, down to one command

Standing up a full local environment for the self checkout stack used to cost an engineer most of a working week. Switching between environment configurations cost several days more. Both now take one command, identically on Linux, macOS and Windows across arm64 and x86_64.

  • 1 week to 1 commandenvironment setup
  • Days to 1 commandswitching configuration
  • Named inventorUS patent pending

The second number turned out to matter more than the first. Reconfiguring was expensive enough that engineers worked inside their own service rather than across the product, so removing that cost let teams treat the platform as one product instead of several.

It has since spread well past engineering. Software architects, team leads, product owners and department managers use it to check the live state of the product without needing to read a cluster.

  • Go
  • Kubernetes
  • Developer experience
  • Patent pending

Selected work

Independent / live product

DreamRunner

An AI interactive fiction platform with paying users, built and operated end to end. Real time generation streamed over Socket.IO, multi provider LLM routing with per model cost accounting and automatic failover, image generation behind a swappable adapter, and text to speech across several engines including zero shot voice cloning served from GPU endpoints.

The architecture is deliberately local first. Everything a user creates lives only in their own browser and is never persisted server side, so the privacy guarantee is a property of the design rather than a promise in a policy. It runs on Kubernetes with billing, 17 localized languages, and full end to end test coverage in CI.

  • Nuxt
  • TypeScript
  • Socket.IO
  • Postgres
  • Kubernetes
  • Stripe
  • LLM routing

Open source

Voice and speech infrastructure

A FastAPI service that exposes Qwen3 text to speech voice cloning on RunPod compatible GPU endpoints, along with containerized tooling for diffusion model training and image generation pipelines. Most of what I know about serving models under real latency budgets came from this.

  • Python
  • FastAPI
  • RunPod
  • Docker
  • CUDA
View on GitHub

How I work

Friction is the real cost

The expensive problems are rarely hard ones. They are the setup step everyone accepted, the process that grew an exception at a time, the bottleneck sitting between two teams so it belongs to neither. I go looking for those.

Tools over output

On AI I would rather point a model at writing the deterministic tool that produces a result than have it produce the result directly. You keep the speed and you get the same answer every time, which is what matters once something runs nightly and nobody is watching.

Owning the whole stack

I have run the database, the deploy, the billing and the frontend of the same product. Knowing what breaks at each layer changes the decisions you make at every other one.

Experience