Higher Self Systems Engineering
Menu

Received a letter from me?

Open the free two-numbers check

AI agents. Integrations. Human accountability.

Your business runs between the tools.

I investigate broken handoffs and build the AI workflows, integrations and checks that move the work forward. One engineer. Clear approval points. Systems your business controls.

A couple of sentences is enough. You do not need a technical brief.

From an investigation / 01

The form looked successful.
The submission wasn’t.

What the visitor saw
“Sent”
What the server returned
Rejected · 403

A sign-up form showed success while its server rejected the request. Checking the handoff exposed what the interface did not.

A finding from a UK wellness engagement. Typographic reconstruction, not a product screenshot.

One named builder: David Bentley

Agreed scope and approval

Your accounts and source code

A documented handover

01 / Selected work

Find the problem.
Build the useful part.

Different tasks, the same discipline: understand the business decision, follow the information and make the work checkable.

Recovery · UK wellness brand

2,151 profiles recovered and verified.

Shopify showed thousands of profiles marked for email marketing. Klaviyo could reach about 150. I traced the gap to an account-wide suppression event, then recovered eligible profiles and verified their status server-side.

Discuss a workflow discrepancy

Supervised AI email workflow

Campaigns prepared.
The founder decides.

AI supports the commercial thinking, writes the emails and builds campaigns inside Klaviyo through its API. I direct the working sessions. Every send waits for the founder’s approval.

Describe a task for an agent

Live tools · Sofema

Tools for a training decision.

A Course Finder and Training ROI Calculator built for Sofema Aviation Services and deployed on its own site. Two practical tools, separate from the knowledge system still in development.

Visit Sofema’s tools page

The diagrams summarise the work; they are not client screenshots. More selected work includes public-source reviews, knowledge systems in development and offline communication research. Explore the work.

02 / What I build

Useful AI.
Joined-up work.

Start with a specific task, not an ambition to automate everything. Make the boundaries clear before handing work to software.

01 / Connect

Integrations and checks

Connect the tools your team already uses. Repair an agreed handoff or build checks around the records that should agree, including a named person to tell when they do not.

02 / Prepare

AI workflows with approval

Draft from approved sources, prepare a campaign or support enquiry triage. Give the agent a defined job, limited permissions and an explicit human approval point.

03 / Keep

A working record the team owns

Bring decisions, procedures, evidence and approved AI work into a shared record. The next person should be able to see what was decided, why and what happens next.

03 / Ways to start

One useful piece
of work is enough.

A known build does not need an audit first. When the problem is unclear, choose an investigation that can actually answer the question.

You know the job

Scope a build directly.

Tell me what should happen, which tools are involved and who makes the decision. We agree a bounded build and how we will check it.

Describe the build

Something inside is not working

Inspect one workflow.

A read-only investigation across the systems we agree. Evidence, a reproduction path and clear limits. Implementation is a separate scope.

Describe the problem

Begin without account access

Review a public surface.

A fixed-scope outside-in analysis of an agreed public part of the business. It establishes what can be checked from outside, not private account settings.

Read the analysis and terms

Scope and price are agreed in writing before paid work starts. See the published pricing and terms. Already speaking to me on Upwork? Keep our conversation there.

04 / Ownership is part of the design

The work stays
with your business.

  • Your accounts and infrastructure.Access is issued by you, limited to the agreed job and revocable.
  • A named human at the boundary.Approval points are designed around actions that change business records or reach customers.
  • Source code, documentation and handover.An agreed check that another engineer can understand the system and take it forward.

Where customer data is processed by an AI model, my delivery policy is an open-weight model on private infrastructure in a region you choose. Messaging channels and integrations have separate data paths, agreed for the engagement.

Third-party services, hosting, maintenance and support are identified in the scope.

David Bentley

Founder and engineer.
One person accountable for delivery.

Expert-Vetted on Upwork

The person doing the work

Business understanding.
Engineering accountability.

I spent years building marketing funnels before moving deeper into software and AI engineering. I look at how customers move through a business as well as how information moves through its software.

That matters when a technically successful automation can still produce the wrong business result.

A conversation, not a technical brief

What should
happen next?

Tell me the workflow you want to improve, the tools involved and what should happen next. A couple of sentences is enough.

Describe one workflow

No paid consultation is required for an initial fit-check message. Any paid work is scoped and priced in writing before it starts.