01
“Our staff spend hours processing the same documents.”
Invoices, applications, quotations and forms retyped by hand into systems that already exist.
TIME LOSTRECURRING
AI systems built around your business.
I build practical AI and automation systems around your existing processes, data and software — with reliability, human oversight and long-term maintainability in mind.
- 4+
- years of professional software engineering experience
- 1
- person between you and the build
- Your
- accounts and infrastructure, wherever practical
Quotation request — matched to accountQuotation request — matched to accountRequired details extracted and validated
VALIDATEDVALIDATEDInvoice batch — documents processingInvoice batch — documents processingReading and checking line items
PROCESSINGPROCESSINGRefund above the approval thresholdRefund above thresholdAbove approval threshold
HUMAN REVIEWREVIEWSupplier enquiry — routed to operationsSupplier enquiry — routed to operationsCRM updated · task created
COMPLETECOMPLETEBuilt to production standards
- Human approval where it matters
- Permission-aware access
- Full audit trail
- Your infrastructure, your accounts
Sound familiar?
The problems usually look like this.
Not “we need AI”. A specific, repeated task that quietly costs a few hours every week and nobody has had time to fix.
02
“Every enquiry is read, classified and routed by a person.”
Someone triages the inbox each morning, decides who owns it, and copies the details into the CRM.
TOUCHPOINTSSEVERAL
03
“Employees constantly search for information we already have.”
Policies, procedures and past projects spread across SharePoint, Drive, email and people’s memory.
SCATTERED ACROSS
SharePointDriveEmail
What better looks like
The same process, with the repetitive part engineered away.
TODAYRepeated manual handling
- Email arrives
- Employee reads it
- Copies the details out
- Checks the CRM
- Creates a task
- Replies manually
WITH AUTOMATIONOne focused review
- Email arrives
- AI understands the requestCLASSIFIED
- Data extracted and validatedVALIDATED
- CRM and task system updated
- A person approvesAPPROVAL
- Response preparedCOMPLETE
TODAYManual handling
- Email arrives
- Employee reads it
- Copies details · checks CRM
- Creates task · replies manually
WITH AUTOMATIONOne review
- Email arrives
- AI understands the requestCLASSIFIED
- Data extracted and validatedVALIDATED
- CRM and task system updated
- A person approvesAPPROVAL
- Response preparedCOMPLETE
What I build
Five services. One of them is where we start.
If ordinary software solves it better than AI, I will say so in the first conversation.
01AI Process Discovery & Solution DesignStart here — turn a business problem into a realistic plan.02AI Workflow Automation & Agentic SystemsSystems that carry out a process, not just answer questions.03Internal AI Assistants & Knowledge SystemsAnswers from your approved documents, with sources and permissions.04Document & Data IntelligenceUnstructured documents and email turned into structured business data.05AI Integrations & Custom Internal ToolsAI inside the software your team already uses, not another chatbot.
01 · START HEREAI Process Discovery & Solution Design
Turn a business problem into a realistic implementation plan, with a written proposal and fixed scope.
02AI Workflow Automation & Agentic Systems
Systems that carry out a process, not just answer questions.
03Internal AI Assistants & Knowledge Systems
Answers from your approved documents, with sources and permissions.
04Document & Data Intelligence
Unstructured documents and email turned into structured business data.
05AI Integrations & Custom Internal Tools
AI inside the software your team already uses, not another chatbot.
Inside a system
AI does the reading. People keep the decisions that matter.
Work that clears every check runs straight through. Anything unusual, expensive or low-confidence stops and waits for a person — with the reasoning and the source document attached.
- Confidence thresholdsThe system knows what it is unsure about.
- Approval thresholdsValue, risk and exception rules you define.
- Full audit trailEvery action, input and approver recorded.
Purchase order reviewSupplier record matched · source attached
OVER THRESHOLDAll line items matchedVAT verifiedPrice differs from last order
Contract renewal — clause changedClause change detected in section 7
REVIEWExpense batch — duplicate checkPotential duplicates need confirmation
QUEUEDReliability
Nothing runs without controls around it.
PermissionsOnly what the user may see
ValidationRules checked before action
Error handlingRetries, then a person
CONTROLLED EXECUTIONAI workflow
Reads, decides, acts — inside the boundaries set for it
Human reviewWhere the decision carries weight
Audit trailEvery step recorded and reviewable
MonitoringAlerts when behaviour drifts
EvaluationAccuracy measured, not assumed
LoggingTraceable from input to outcome
Selected work
A few projects, explained as problems.
MON
TUE
WED
THU
AI-powered meal planning product
Structured generation with hard constraints — the interesting engineering is making a model produce something valid every single time, not producing text.Structured generation with hard constraints, producing a valid plan every time.
Applied computer-vision research
Where the model is confident enough to act alone, and where it should defer — the same question every business automation has to answer.Where a model is confident enough to act alone, and where it should defer.
Interface visuals are conceptual, not product screenshots. No unapproved client identity or outcome is shown.
How I work
A fixed-scope project, start to handover.
You buy an outcome, not engineering hours. Well-defined projects are priced as a fixed scope with a written proposal before anything is built.
Understand
FREEYou describe the process as it works today. I look for where the time actually goes and whether AI is the right tool at all.
Design & propose
1–2 WEEKSSolution, integrations, security, approval points, phases, timeline and a fixed price — in writing, before commitment.
Build
ITERATIVEBuilt in visible increments with your feedback, against real examples from your business rather than sample data.
Test & deploy
Accuracy measured against your own cases, then deployed into your infrastructure and your accounts.
Handover
OPTIONAL SUPPORT AFTERCode, documentation and control transferred to you. Ongoing support is available, but never a condition of the system running.
Understand
FREEYou describe the process as it works today. I look for where the time actually goes.
Design & propose
1–2 WEEKSSolution, integrations, approval points and a fixed price, in writing.
Build & test
Visible increments, measured against real examples from your business.
Deploy & hand over
Into your infrastructure and your accounts. Support is optional, never a condition.
Direct access
The person you brief is the person who builds it.
TYPICAL CONSULTANCY
ClientSalespersonAccount managerProject managerConsultantTechnical leadDeveloper
WORKING WITH ME
ClientLuke
Fewer layers, faster decisions, and small projects that stay commercially viable.
Ownership
Your systems stay yours.
I don't host your production systems or hold your accounts hostage. Everything is deployed into infrastructure you control.Everything is deployed into infrastructure you control.
Cloud accounts & infrastructureCloud accountsYOURS
Source code & repositoriesSource codeYOURS
Databases & company dataDatabases & dataYOURS
AI provider accountsAI provider accountsYOURS
Technical
credibility
AI
- LLM integration
- Agents
- RAG
- Computer vision
- Evaluation
Backend
- .NET
- Python
- Node.js
- REST APIs
- SQL
Frontend
- Angular
- React / Next.js
- TypeScript
Cloud
- Azure
- Docker
- CI/CD
- Linux
Describe how the process works today.
Where does it start, who touches it, and where does it slow down? If AI isn't the right answer, I'll tell you that in the first conversation.