Start a project
OXEN.DEV
SHEET 01 / 01 · SERVICES · CAPABILITY SCHEDULE

We train the models that run your business.

Oxen trains custom models, builds the AI systems around them, and operates the data infrastructure underneath. We work on retainer for a small number of clients at a time, and we run our own products on the same practices. Every capability below names something we have shipped.
We reply within1 business day
Working proof of concept in2 weeks
New models in production within2 days
Uptime on systems we operate99.9%

What we do

04
01

Custom models and machine learning

Models trained on your data, not a prompt wrapped around somebody else's.

We train models rather than only calling them: autoregressive forecasting on your own time series, image models for classification and detection, facial recognition, and language models built or fine-tuned for a single domain. That includes the unglamorous half, which is where most projects fail. Labelling, feature engineering, reproducible training runs, an evaluation harness that catches regressions before your users do, and drift monitoring once the model is live.

  • Autoregressive forecasting on your time series
  • Image models: classification, detection, and segmentation
  • Facial recognition and identity matching
  • Custom and fine-tuned language models for your domain
  • Analysis and feature engineering over large datasets
  • Reproducible training pipelines
  • Evaluation harnesses and drift monitoring
  • On-device deployment with Core ML and Metal

Machine learning we have shipped to real users. These native apps run pre-trained models on device, which is the deployment half of the same practice:

02

Applied AI systems

Agent workflows that do a real job, under supervision, and measured against the process they replaced.

We build for the daily grind of a business: document intake, review queues, customer response, and the handoffs between them. Every workflow runs with a human in the loop where it matters, logs what it did, and is scored against the manual process it replaced. When a better model ships, our clients are usually running it within two days.

  • Workflow design and scoping against a measured baseline
  • Evaluation harness with regression tests on real cases
  • Human review queues and approval gates
  • Model routing, with fallback and cost controls
  • Tracing, monitoring, and alerting on quality drift
  • Model upgrade path, so new releases ship in days
03

Data infrastructure

The load-bearing layer underneath the product, built to be operated rather than handed over and forgotten.

APIs, pipelines, multi-tenant backends, authentication, and the migrations that get you from what you have to what you need. We run this layer ourselves across our own products, so the practices are ones we live with rather than ones we recommend.

  • Multi-tenant schema design and migrations
  • REST and MCP surfaces, with generated API documentation
  • Authentication, scoped keys, and OAuth for connected apps
  • Ingest and transformation pipelines
  • Observability: logs, traces, and paging that reaches a person
  • Backup, restore, and a tested recovery procedure

The workspace we run on this stack, with a REST v1 API, an MCP server, and OAuth 2.1 for connected apps:

04

Product engineering

Web and native applications taken from a sketch to something customers pay for.

We design and build the whole product: the interface, the backend behind it, billing, and the operator tooling that makes it supportable. The products in our own register were built this way, which is why the estimates we give you are drawn from work we have actually finished.

  • Product and interface design
  • Web applications on Next.js and React
  • Native apps for macOS, iOS, and iPadOS
  • Billing, subscriptions, and customer accounts
  • Operator consoles and internal tooling
  • App Store submission and release automation

Products we designed, built, and now operate with paying customers on them:

How we work

04
  1. 01DAY 1

    Scope

    You send a note. We come back within one business day with questions and a straight answer on whether we are the right firm for it. If we are not, we say so.

  2. 02WEEK 1

    Written scope

    Before anyone writes code you get a written scope: what we are building, what it costs, what it does not include, and how we will know it worked.

  3. 03WEEK 2

    Proof of concept

    A working proof of concept, running on your data, usually lands inside two weeks. You decide whether to continue on the strength of something real rather than a deck.

  4. 04ONGOING

    Build and operate

    Two-week cycles with a demo at the end of each. When the system goes live we keep operating it: monitoring, on-call, and model upgrades as they ship.

How to engage us

03

Discovery sprint

For teams who know the problem, not the shape

Two weeks, fixed fee

For teams who know the problem but not the shape of the solution. Ends with a working proof of concept and a written build plan you own, whether or not you continue with us.

  • Baseline measurement of the current process
  • Working proof of concept on your data
  • Written build plan and cost estimate
  • No obligation to continue

Build engagement

For taking a proof of concept to production

Monthly, minimum three months

A dedicated team taking a system from proof of concept to production. Two-week cycles, a demo at the end of each, and a scope you can change between cycles.

  • Dedicated engineering team
  • Two-week cycles with a demo each cycle
  • Direct access to the people doing the work
  • Scope adjustable between cycles

Operate

For keeping a live system healthy

Monthly, rolling

We keep running what we built. Monitoring, incident response, and model upgrades, on the same infrastructure practices we use for our own products.

  • Monitoring and alerting
  • Incident response with a named contact
  • Model upgrades as new releases ship
  • Quarterly review against the original baseline

Every engagement is scoped to the problem. We have run two-week pilots and multi-year programs, and we will tell you which one you need in the first reply.

Tell us what you're building.

Send a note about the problem and where it stands. We reply within one business day with questions and a straight answer on whether we should take it on.

Start a projectSee the work