Independent consulting
Intrepid Solutions is the independent practice of Cedric Keown — working across business process outsourcing, systems architecture and applied AI for organisations that need the thinking and the build, not a deck.
Three disciplines that rarely stay in their lanes. Most engagements touch at least two.
Operating-model design, vendor selection, transition and governance. The unglamorous part: who does what, how it gets measured, what the service really costs, and what happens on the day it goes wrong.
Integration, platform and data design for systems carrying real operational load. I favour architectures a small team can still run in three years' time over ones that look impressive on a slide.
Where AI genuinely pays, and how to put it into live operations without losing control of cost, quality or accountability. I run this in my own delivery: autonomous agents that implement written work orders and merge without a human reviewer — because the review is done by independent machine gates and the repository's own tests, and every run is costed, logged and auditable. A model with commit rights is a liability until the controls around it are explicit and measured. Then it is capacity.
Short engagements, senior hands, no pyramid. You get the person you met.
A standing call on my time. For teams who need architectural judgement and a second opinion on the record, without carrying the cost of a permanent hire.
A defined problem, a defined deliverable, a defined price. Assessments, target operating models, architecture reviews, build.
Embedded with your team for an agreed slice of the week — designing, reviewing and building alongside the people who will own it afterwards.
A sample of recent engagements. Client names available on request.
The problem. A small development team supporting a growing portfolio of production portals, with more backlog than hands and no appetite for unsupervised AI touching production code.
What I did. Designed and built a platform that takes a written work order, implements it inside the target repository's own guardrails, and merges only when that repository's gates pass. Independent review gates sit in front of and behind the work; every run carries a full event log, cost and token accounting, and an audit trail of who approved what.
The controls are claims to be tested, not articles of faith. Every blocked run is scored — fair block or false alarm — and a new check runs advisory, recorded but not blocking, until the evidence says it has earned the right to say no.
The outcome. Routine, well-specified change now moves from written work order to merged code without a human reading the diff — and the controls that make that defensible are measured, not assumed.
The problem. Client data subject to POPIA had to remain in South Africa. The managed database platform in use had no South African region, and re-platforming the applications was not viable.
What I did. Designed a self-hosted equivalent of the managed stack running in AWS af-south-1, reduced to a single provisioning command — private networking with no inbound SSH, session-manager-only access, automated backups, and alarms on the paths that matter.
The outcome. Applications kept their existing data layer and API surface while the data itself stopped leaving the country. The same toolkit now stands up further in-country environments in minutes rather than weeks.
The problem. Client-facing applications needed data from a legacy policy administration platform, but each client site's firewall whitelisted only specific origins. Direct integrations worked in development and failed in production.
What I did. Consolidated every integration onto a single brokered service layer holding the credentials server-side, with one whitelisted egress path and a uniform contract for the applications in front of it.
The outcome. New integrations became a configuration change rather than a firewall negotiation, and credentials left the client applications entirely.
The problem. Weekly client billing depended on someone manually collating time records across projects — slow, late, and inconsistent enough to cost real revenue at invoice time.
What I did. Designed and implemented an automated cycle that draws time records directly from the tracking system, applies the commercial rules per project and work type, and issues the report to its recipients on schedule — with every run logged and re-runnable for a given period.
The outcome. The reporting week closes on time without manual effort, and disputes are settled against a recorded, reproducible extract.
I'm Cedric Keown. I work with organisations in financial services, insurance and operations-heavy industries — usually where a business problem and a technical one have become the same problem.
My work sits between the operating model and the architecture. I've designed the processes and picked the vendors; I've also written the code, run the deployments and carried the pager. That combination is the point: recommendations that account for what it actually takes to run the thing.
Intrepid Solutions is how I engage independently, including ongoing work with CustomApp and Cardinal Systems.
If you have a problem that spans process and platform, I'm interested in hearing it — including when the answer turns out to be that you don't need me.