How we work
AI product development and engineering for other companies
Most AI projects stall between the prototype and production. Retrieval returns the wrong passage, nobody has an eval set, cost per request is unknown, the model has no audited path into the systems it needs, and the first model update breaks a flow nobody tested. We design, build and ship the production system: agents, LLM applications, RAG, evals, integrations and native mobile apps, in your repository and your cloud.
Running software as the deliverable
No strategy deck. The output is a deployed system with tests, evals and runbooks, in a repository with your name on it.
Patterns proven on our own traffic
We run our own products on the same stack. Retrieval, eval and agent patterns reach clients after they have held up under real load.
Supplier only
We build for your market, never in it, and decline work that would put us opposite a client in their own category.
Client ownership of every asset
Repositories, cloud accounts, model provider keys, data and domains are yours from the first commit. Fire us on a Friday, ship on Monday.
Process
Four phases, written into the contract
- 01
Scope
One week, written, fixed shape
One working session, then a scope document: what ships, what does not, the architecture, the data it needs, how it will be evaluated, what it costs. No sixty-page discovery invoice.
- + Scope doc
- + Architecture and data map
- + Fixed price or rate plus ceiling
- 02
Thin slice
Production by week three, narrow
One real path end to end: auth, model calls, tracing, cost tracking and the deploy pipeline. Everything after is widening a road that already exists.
- + Live URL
- + CI on push
- + Cost per request model
- 03
Widen
Weekly increments behind the eval harness
Features land behind evals and tracing. A change that regresses accuracy, latency or cost does not merge.
- + Eval suite
- + Traces and dashboards
- + Weekly one-page readout
- 04
Hand over
Or keep us on call, both are normal
Runbooks, architecture notes, a working local environment, the eval suite and a walkthrough with whoever inherits it. The exit is written into the agreement on day one.
- + Runbooks
- + Local env
- + Named owner
House rules
What we hold ourselves to
- 01Running software over adviceThe deliverable is a deployed system, not a recommendation about one.
- 02Evals before featuresEvery AI behavior has a test set. A change that regresses it does not ship.
- 03Client ownershipRepositories, cloud accounts, model keys and data in your name. Nothing held hostage.
- 04No competing with clientsWe build for your market, never in it, and decline work that would place us opposite you.
- 05An invisible supplierYour product carries your brand. Ours stays off the work.
Engagement models
Build
Fixed scope. We design, build and ship it; you own the output outright and run it yourself.
You know what you need built.
Operate
Retained. We run and stay on call for what we built, reported against eval, uptime and cost numbers.
You need it built and kept running.
Partner
Equity or revenue share on a product we build with you. Selective, and we say no often.
You want an engineering partner, not a vendor.
Frequently asked questions
Something else? Email t@moher.ai and an engineer will answer it.
What does Moher do?
Moher is an AI development and engineering company. We design, build and ship AI products for other companies: agents, LLM applications, RAG and search, eval harnesses, integrations into existing systems, and native mobile apps. The deliverable is running software in your repository.
How is Moher different from other AI development companies?
We put the boring parts first. Every build ships with an eval harness in CI, a cost-per-request model, traces, and runbooks, and it runs in your cloud accounts from the first commit. Most AI projects fail on retrieval, evaluation and cost, not on the idea. That layer is our daily work.
How fast can you start?
Scope in one week. A thin slice of the product is usually live in production by week three, then we widen it in weekly increments measured against the eval suite.
Who owns the code?
You do, from the first commit. Repositories, cloud accounts, model provider keys and domains are in your name. You could end the engagement on a Friday and ship on Monday.
Which models and platforms do you build on?
Claude, GPT, Gemini and open-weight models, chosen per task on cost, latency and eval results. Typical stacks are TypeScript or Python, Postgres with pgvector, Vercel or AWS, and Expo for native mobile.
What does an AI development project cost?
A scoped first build typically runs $40,000 to $150,000 depending on integrations, data work and how many surfaces ship. Retained engineering is a monthly rate. You get a fixed price or a rate with a ceiling in the scope document before any work starts.
Do you build mobile apps?
Yes. Native iOS and Android through both app stores, sharing a codebase with web where the product allows it.
Do you compete with your clients?
No. We build for your market, never in it, and we decline engagements that would place us opposite a client in their own category.
Tell us what you need built
Four quick questions and an email. An engineer reads every brief and replies within one business day with next steps, and a written scope follows within a week.
- Written scope and price within a week
- Production deploy by week three
- Your repo and cloud from the first commit
- Senior engineers only, no account managers
Start a project
Step 1 of 5
Got it. An engineer is reading your brief.
You will hear from us within one business day, usually sooner. If it is urgent, email t@moher.ai.