In the latest of many Barefoot projects I've been revitalising the platform of a major US legaltech firm. Starting with thorough analysis, the project advanced to a range of technical and procedural measures designed to enhance resilience, improve developer experience and bring the system up to SOC2 compliance.I've also unlocked a significant improvement in customer experience by working with developers to make applications play better with the underlying platform - a balance often overlooked in the early stage of startups. This enabled a shift to autoscaled compute in the Kubernetes clusters, greatly increasing the throughput of the system when it's needed yet reducing overall cost by shedding capacity when it isn't.

Work at a couple of clients through the well-respected US-based Barefoot Coders consultancy. Projects included rearchitecting build & infrastructure automation systems for large data science platforms running on Google Cloud. The most notable project was writing a robust Terraform provider for deploying & configuring Google's "Looker" analytics service. Written in Go, this translated a large amount of Looker's API surface area into Terraform resources and thus enabled a fully automated, joined-up integration of the service with the client's existing infrastructure. The result was a reliable platform with much of the complexity abstracted away, needing trivial effort to replicate for the user's new clients.

DevOps & Python consulting through my own company for a variety of clients.

OneFlow (first a startup but quickly acquired by HP) is a cloud-based print preprocessing system for books, photos, mouse mats, mugs, clothing and just about any object you can imagine. The service is offered to print houses to prepare and optimise millions of documents a day. With 50+ processing steps that can be arbitrarily rearranged by the customer, on documents as large as 50 GB, the organisation needed a huge platform on which to run them. With hundreds of print houses around Europe depending upon the system to keep their room-sized printing presses moving, resilience and performance were considered paramount.My team and I kept outdated Kubernetes clusters on life support while building an autoscaling replacement based upon AWS EKS. Wholly managed by Terraform and a handful of hand-rolled Python scripts, the new system featured autoscaling using AWS spot instances for a huge cost saving against its predecessor, greatly improved monitoring capabilities which integrated knowledge of the infrastructure with that of the apps it hosts, and far more transparency to OneFlow’s developers on how their applications were performing. Terraformed infrastructure also made it easy to maintain identical dev/staging platforms for the developers and with the code stored in a GitHub repo it was much easier to educate the teams about how systems were working under the hood.Once tested and launched we slowly migrated workloads onto it, realising a 50% cost saving on the previous system and significant increases in reliability and throughput.As DevOps Lead, as well as working closely with the CTO and Development Leads to come up with a vision for the architecture I also bore responsibility for hiring, day-to-day workload allocation, infrastructure budgeting / capacity planning and outage post-mortems. Given the sheer scale, complexity, throughput and importance of the stack the job was a real challenge with plenty of notable successes along the way.

Re-hosted a large microservice node.js + RethinkDB web application on Google Cloud. The resulting stack (serving three news brands) is deployed to GKE with a set of Helm charts and a minimal amount of custom tooling for CDN & image serving. GKE has turned out to be low-maintenance and rock solid; I'll be glad to work with it again.

Rewrote a number of complex Python applications geared toward processing DNA samples to extract information on a subject’s ancestry. Given the huge volumes of data involved I constructed (with Terraform for config management) an AWS Batch cluster to pipeline jobs and run them on the cheapest AWS instances available.Improvements to parallelization within the app and resource selection (particularly AWS spot instances) outside it have led to a significant cost saving for the client.