ENG-012 weeks
Discovery sprint
A fixed-scope fortnight to rank use cases, audit your data, measure candidate models and size the hardware. It ends with a written plan, a configuration sheet and a quote for the build.
CSTM-DEV · REV AcstmAI™ · Development
Custom AI development, from discovery to handover.
Our engineers build the agents, applications and integrations that make the hardware and software earn their keep, measure them against your own cases, and hand you everything they wrote.
Fig. 01 · Wiring a motor controller, NASA JPL
What we build
Five kinds of custom AI work.
Each starts the same way: one workflow, one evaluation set, and a clear definition of done.
Custom AI agents
Agents that do one job inside your systems, tested on your cases
Read more →
Custom AI applications
Copilots, review tools and internal apps shaped around the task
Read more →
Document & vision AI
Extraction, classification and checking for scans, forms and drawings
Read more →
Fine-tuning & model training
Open-weight models adapted to your terms, formats and tasks
Read more →
Systems integration
AI connected to ERP, CRM, documents and data, with scoped access
Read more →DEV-06 · PROCESS
How a build runs
Discovery, fixed-scope build, burn-in, install and operations.
Read more →How we work
Four ways to engage.
Most clients start with discovery, then a fixed-scope build. Co-build and managed operations fit around your own team.
ENG-02Per phase
Fixed-scope build
Agreed deliverables, built against an evaluation set and demonstrated weekly. Scope changes are written down and re-quoted, not absorbed silently.
ENG-03Flexible
Co-build with your team
Our engineers work alongside yours in your repositories, so your team learns the stack while it ships, and owns it afterward.
ENG-04Ongoing
Managed operations
After launch we monitor, update models, re-run evaluations and handle hardware issues, for teams who would rather not staff it.
Build process
From first call to running in your rack.
Each phase ends with a review and a go/no-go. A first system is typically in production 8–12 weeks after discovery begins.
012 weeks
Discovery
Pick the workflow, audit the data, size the hardware, write the evaluation set.
- Use-case ranking
- Data audit
- Configuration sheet
- Quote
024–7 weeks
Build
Software built and scored against the evaluation set, reviewed weekly with your team.
- Weekly demos
- Eval scores
- Integrations
03≈ 1 week
Burn-in
Hardware assembled and run under sustained load on our bench, with the software loaded.
- Stress tests
- Imaging
- Acceptance run
041–2 weeks
Install
Racked, cabled, networked and joined to your identity provider, on your site.
- Site install
- SSO + roles
- Handover
05Ongoing
Operate
Your team runs it with our runbooks, or we run it for you. Models updated as they improve.
- Monitoring
- Model updates
- Quarterly evals
Typical first system: 8–12 weeks to productionThe full process →
Reel 02 · Five-axis CNC milling, NASA Kennedy1998
FAQ
Working with our engineers.
Do we have to buy cstmAI hardware to work with your developers?
Our builds are designed, tested and supported on cstmAI systems, which is how we can answer for the whole result. If you already run suitable hardware, tell us in discovery and we'll assess it.
Who is on the team?
Engineers, a product lead and, for applications, a product designer, working with the people in your business who know the workflow. You'll meet the engineers on the first call.
What do we get at handover?
The code, prompts, evaluation sets, integration documentation, runbooks and any model weights we trained, in your repositories and storage.
Get a quote
Spec your system.
Tell us the models you want to run, how many people will use them and where the hardware should live. An engineer replies with a first configuration and the questions that decide the quote.