Applied AI
AI that does one specific job well, measured against what it costs.
Most AI projects fail in the same place: a convincing demo that nobody can put in front of customers, because nobody measured whether it is right often enough, or what it costs at volume.
- From
- $1,800
- Typical timeline
- 2 weeks
- Stack
- Python · TypeScript · PostgreSQL · Vector search · OpenAI API
We start with a feasibility sprint. You get a working prototype on your own sample data, an evaluation set of real examples with measured accuracy, and a running-cost estimate at your volume — before committing to a build.
Only after that do we offer a pilot build: one use case inside your product, with acceptance criteria, fallback behaviour when the system is unsure, and monitoring.
What this covers
Search and Q&A over your own content
Prototype: ask questions of your documents, policies or product data and get answers with citations back to the source.
Structured extraction
Prototype: turning PDFs, emails and forms into clean, validated data — invoices, applications, reports.
Triage and routing
Prototype: classifying and routing incoming requests or documents so the right person sees the right thing first.
Measured accuracy
A test set of at least 30 real examples, each checked by someone who can judge the answer. You get a number, not an impression.
Running-cost model
What the feature costs at your projected volume — treated as a design constraint from day one, not a surprise on your first invoice.
Privacy and provider review
What data goes to which provider, in writing, before any of your data touches a service.
What you get
- A working prototype on your sample data
- An evaluation set with measured accuracy and a failure analysis
- A running-cost estimate at your volume
- A written recommendation with effort and price for a pilot
- For pilot builds: acceptance criteria, fallbacks, monitoring and cost alerts
Stack
- Python
- TypeScript
- PostgreSQL
- Vector search
- OpenAI API
- Anthropic API
We use a small, well-understood stack deliberately. Fewer surprises, faster decisions, and estimates that hold.
Common questions
- Will our data be used to train a model?
- No. We use commercial API tiers that do not train on submitted data, and we will confirm the specific provider terms in writing before any of your data touches the service.
- How do we know it is accurate enough?
- We build an evaluation set from your real cases and measure against it before we say a feature is ready. If the number is not good enough, we say so.
- What does it cost to run?
- You get a cost model at your projected volume before committing to the build. Running cost is a design constraint from day one.
- Can you build a full AI feature on day one?
- No — and we would not advise it. We start with a feasibility sprint, then a pilot build, then scale only when the evaluation and costs justify it.