About Gem State Digital

Hi, I'm John, founder of Gem State Digital. We're an AI software consultancy based in Boise, Idaho. We help organizations design, build, and ship AI systems that hold up in production — and we're honest about what's production-ready and what isn't.

That honesty is the whole point. A lot of AI work today is slideware: impressive demos that quietly fall apart the moment they meet real data, real users, and real operational load. We come at it from the other direction. We build the system first, run it, learn where it breaks, and only then tell you what it can do.

Our mission

Help organizations ship their first AI workflows to production — and keep them running. We pair modern AI engineering with the unglamorous discipline that makes software last: type safety, tests, infrastructure as code, and architectures that stay understandable as load grows.

How we work: AI-first, AI-augmented

We're an AI-first shop in two senses.

First, AI is a core part of what we build — agent design, multi-agent orchestration, RAG pipelines, and the production-safe patterns that keep all of it on track.

Second, AI is part of how we build. Our engineers are AI-augmented, which lets us deliver more, faster, and at lower cost. That efficiency is what makes our second offering possible.

So there are two distinct things we do, and we keep them separate:

  • AI consulting. Agent design, orchestration, and production-safe AI delivery — getting LLMs to behave, perform, and stay affordable in the field.
  • Affordable traditional and legacy software. Backend services, web apps, and the boring-but-critical systems your business actually runs on — built faster and cheaper by AI-augmented engineers.

Both offerings are genuine. Sometimes the right answer to your problem is a sophisticated multi-agent system. Sometimes it's a well-built API and a static site. We'll tell you which, even when it's the less exciting one. When you can build a proper state graph, you should — and when you don't need one, we won't sell you one.

What we build

AI and agentic engineering — get LLMs doing useful work without the surprises.

  • AI agent design and multi-agent orchestration with LangChain and LangGraph
  • Production-safe delivery: tool-use patterns, evaluation, and behavior-steering to reduce hallucination, improve performance, and control cost
  • RAG pipelines and LLM output and truthfulness evaluation
  • Getting LLMs to safely use data you don't hand them directly

Backend and cloud — the durable services your product depends on.

  • Python backend development with FastAPI, async, and Pydantic
  • AWS serverless architecture — Lambda, API Gateway, S3, DynamoDB
  • Infrastructure as code with Terraform and Terragrunt, reusable modules, and OIDC-based CI/CD
  • Java and Spring Boot for enterprise and legacy systems

Delivery and engineering practice — the habits that keep software maintainable.

  • Type-safe API client design with high test coverage
  • Full-stack and static web delivery with Next.js and TypeScript on AWS
  • DevOps automation with GitHub Actions, Jenkins, and Docker

The work behind the words

We'd rather show you than tell you. Our public projects are work you can actually inspect:

  • A multi-agent framework with persistent per-agent workspaces, custom tools, and scheduled tasks, built on LangGraph
  • An original LangGraph behavior-steering middleware pattern that keeps long-running agents on track at runtime
  • An LLM model router that directs queries to simple or advanced models via conditional routing
  • A multi-agent system that generates 3D-printable objects — AI applied to a hardware domain
  • A multi-model truthfulness evaluator with evidence gathering
  • A type-safe Python API client with full validation and 99% test coverage
  • Live production sites on AWS S3 and CloudFront with GitHub Actions OIDC CI/CD
  • A reusable, production-ready Terraform module library for AWS

We publish our engineering notes and hands-on experiments too. When something is a demo, we say so. When something is running in production, we say that.

The founder

Gem State Digital is led by John Sosoka, who has delivered high-throughput software to Fortune 500 companies. His career runs from software tester to principal engineer and tech lead, with consulting work across legal software, banking, e-commerce, grocery, and construction. Along the way, he's helped organizations ship their first AI workflows to production.

That arc matters. Testing teaches you how software actually fails. Leading teaches you how to keep a system understandable as it grows. Both habits show up in everything we build.

What makes us different

  • Deployed systems, not slideware. We've shipped public work you can inspect, and we're explicit about what's production-ready and what's a demo.
  • The right tool, not the trendiest one. Sometimes that means a serious AI system. Sometimes it means we talk you out of one. We optimize for what works and what you can maintain.
  • Human-in-the-loop by design. We build semi-autonomous systems with the controls and evaluation that keep them safe — not set-and-forget promises.
  • Direct and senior. You work with the people building the thing — no handoffs, no account-manager layer. No bots, just genuine conversations and engineer-to-engineer honesty.

Is this a fit?

We're a good match if you want to ship your first AI workflow to production, need someone candid about where AI helps and where it doesn't, or want backend and web software built efficiently by AI-augmented engineers. We're probably not the right call if you're shopping for a flashy demo with no intention of running it for real.

Let's talk

The best place to start is a short, low-pressure conversation: tell us the problem, and we'll tell you honestly whether and how we can help — and scope it from there.

Based in Boise, Idaho. From Fortune 500 high-throughput systems to your first production AI workflow — get in touch and let's build something that lasts.