Skip to content

ApproachHow we work

How we think about intelligent systems

Smith Advanced Systems is an independent AI and technology research lab. We study hard problems in agents, applied ML, and the software that has to work under real operational pressure, then turn what we learn into tools and projects.

What we focus on

Research aimed at production systems

We focus on the problems that appear once a system leaves the demo. Automation that has to hold up under load. Models that need durable features, review workflows, and ownership, not just another notebook. Software teams that need AI capabilities without surrendering the stack to a black box. That focus shapes how we choose research directions and what we build.

Research areas

Where the projects draw from

Agentic software systems

Multi-agent orchestration, guardrails, observability, and deployment patterns for workflows that have to survive production. Projects like skailr and olios come from this thread: AI that works in files and structures you own, with every change reviewable.

Applied machine learning

Model-first approaches to matching, prediction, and decision support, with explicit review loops and evaluation. We care about systems that stay operable as inputs drift, not one-off models that only work on yesterday's snapshot.

Intelligent systems integration

How AI components attach to the platforms teams already run: APIs, SDKs, warehouses, and operational services. The goal is software that fits the stack rather than forcing a rebuild around a vendor monolith.

Evaluation and synthetic testbeds

Intelligent systems fail on messy inputs. We build evaluation tools and synthetic datasets (including generate-data) so teams can stress matching, scoring, and agent workflows before production surprises them.

How we build

From idea to operable software

  1. Discovery

    • Study how existing systems behave and fail
    • Map problems and constraints
    • Hypothesize better tools

  2. Implementation

    • Prototype and design architecture
    • Test on realistic workloads
    • Iterate on where the tools fit in

  3. Operationalization

    • Harden for production
    • Document integration patterns
    • Spin out into independent projects
We care about how systems behave years after launch, not just on day one. The projects inside Smith Advanced Systems are built slowly and deliberately, with architectures meant to be owned and evolved by the teams that run them.
Afterword

Want to collaborate around these problems?

If you're working on similar problems in intelligent systems, or interested in the projects in our portfolio, we're open to thoughtful conversations.

Start a conversation