Your documents are not standard
The forms, reports or specifications you process are specific to your trade, and generic extraction tools handle them badly.
04 Custom AI solutions
When off-the-shelf AI does not fit the actual problem.
Some problems are specific to how one business operates — the document nobody else uses, the judgement that depends on your own history, the tool that would only ever make sense to you. That is what custom is for.
- We tell you when a product would be cheaper
- Built on your data, kept in your control
- You own the system and the documentation
The problem
The gap products do not cover.
The AI market is full of good products solving common problems. The difficulty is that the expensive problem in your business is usually not a common one.
The judgement depends on your history
Which jobs are profitable, which clients are worth chasing — the pattern is in your records and nowhere else.
Products cover eighty percent
And the missing twenty percent is the part that actually costs you time, so the tool never gets adopted.
Nothing joins up
Several AI tools bought separately, none aware of the others, each needing its own manual step to be useful.
How it works
Custom AI Solutions, in shape.
01
Your data
Records, documents and history you already hold
02
Preparation
Cleaned, structured and labelled
03
AI capability
Extraction, classification or prediction
04
Your systems
Delivered where the work happens
Our approach
Feasibility first, then the narrowest thing that works.
We start by checking whether it is achievable with the data you have, and whether a product would do it more cheaply. If custom is the answer, we build the narrow piece that removes the cost — not a platform.
-
A feasibility assessment first
An honest answer on whether your data supports it, before anyone commits to a build.
-
Built on your own records
Your history, documents and terminology, rather than a general-purpose model guessing at your domain.
-
Integrated where the work happens
Inside the systems your team already uses, rather than another tab nobody remembers to open.
-
Data handling agreed up front
Where it is processed, what is retained and who can see it — decided before anything is connected.
What's included
What we build.
Scoped after a feasibility assessment, so nothing gets built that a product would have done better.
Feasibility assessment
Whether your data supports the outcome, what accuracy is realistic, and what the alternatives cost.
Custom AI applications
Purpose-built tools for a specific task, with an interface designed for the people using it.
Internal AI assistants
Assistants that answer from your own procedures, records and documentation.
AI document processing
Extraction and classification for the document types specific to your business.
Data preparation
Cleaning, structuring and labelling your records so a model has something reliable to work from.
Model selection and evaluation
Choosing between existing models and comparing them on your data rather than on benchmarks.
AI integrations
Embedding AI capability into software you already run, through its APIs.
Business-specific workflows
Multi-step processes combining AI judgement with deterministic rules.
Accuracy monitoring
Ongoing measurement so drift is caught rather than discovered.
Documentation and handover
Full documentation so the system is maintainable by someone other than us.
Use cases
Where custom earns its cost.
Realistic examples. What fits your business depends on your data and process — that is what the first call establishes.
Trade-specific document processing
Reading the forms, specs or reports unique to your industry, and extracting the fields that matter.
Internal knowledge assistant
Staff query years of procedures, project records and documentation in plain language.
Job profitability prediction
Using your own completed job history to flag which quotes are likely to run over.
Quote and proposal drafting
Assembling drafts from your historical pricing and past documents for a person to review.
Quality checking
Reviewing submissions or records against your own standards before they go out.
Embedded AI features
Adding classification, summarisation or search into software you already own.
Technology
What we build with.
Tools we actually use. If your stack needs something not listed, ask — we will tell you honestly whether we can integrate with it.
AI & models
LLM integrationRetrieval augmentationText classificationDocument extractionComputer vision
Application
Web applicationsInternal toolsREST APIsPHPJavaScript
Data
MySQLPostgreSQLVector searchData pipelines
Infrastructure
AzureAWSSelf-hosted deploymentn8n orchestration
Outcomes
What custom delivers.
It fits the whole problem
No missing twenty percent, so it actually gets adopted.
Built on knowledge only you have
Your history becomes an operating advantage rather than dormant records.
Measurable against your own data
Accuracy evaluated on your cases, not on a vendor's benchmark.
Control over your data
Where it goes and what is retained is your decision, agreed before the build.
Works inside existing systems
Capability added where the work already happens.
A system you own
Source code, documentation and deployment — not a subscription you cannot leave.
How it runs
Feasibility, build, prove, hand over.
Seven stages. Nothing touches live systems until it has been tested against real cases.
01
Discover
The problem, the data you hold, and what a good outcome measurably looks like.
02
Plan
Feasibility assessment, realistic accuracy target, and the alternatives costed.
03
Build
Data preparation, model selection and application development.
04
Integrate
Connected to the systems where the work actually happens.
05
Test
Evaluated on your real historical cases against the agreed target.
06
Launch
Deployed alongside the existing process until accuracy is demonstrated.
07
Improve
Accuracy monitored, retrained or tuned as your business changes.
Questions
About custom ai solutions.
If yours is not here, ask us directly — you will get a straight answer rather than a sales call.
Three tests. The task has to be repetitive enough that automating it saves real time. You need enough historical data for a model to learn the pattern. And no existing product should cover it adequately. Fail any of those and we will tell you to buy something instead — that conversation is free.
It depends entirely on the task. Retrieval-based assistants work from documentation you already have, so often no training data at all. Prediction from your own history needs hundreds of examples at minimum, and more is better. We assess this during feasibility, before you commit.
It usually is. Cleaning and structuring it is a normal part of the project rather than a reason not to start. What matters is whether the information is present at all — formatting we can fix.
That is a decision we make with you before anything is connected. Options range from self-hosted models on your own infrastructure to commercial APIs with defined retention terms. We will tell you the trade-off in cost, accuracy and control for each.
It depends on the task, and we will not quote a number before assessing your data. What we do commit to is measuring it on your real cases and telling you honestly whether it meets the target we agreed — including when it does not.
You do. Source code, documentation, trained artefacts and deployment configuration. You are not required to keep working with us to keep using it.
Related AI solutions
These are usually built together.
Most projects combine two or three — an agent that runs on n8n workflows, or a chatbot that hands to a custom system.
Describe the problem before the technology.
We will tell you whether AI solves it, whether a product already does, and what a build would realistically involve.