Proof of work

Systems we built, running in production.

Record volumes, request loads, and the constraints they were delivered under. Every entry links to a written account of how it was done.

Our clients are not named. The engineering is described in full.

300 million records, zero downtime

Partitioned and migrated a Heroku Postgres estate serving real-time visibility across every US ZIP code, with no outage and no maintenance window.

Delivered for a Fortune 100 telecommunications providerHow we did it

100M+ requests, held under load

Dyno strategy, Redis architecture and failure-mode design for a national real-time data and API platform.

Delivered for a Fortune 100 telecommunications providerHow we did it

AI reliability without touching the model

A Feast-centric feature store that stabilised the full ML lifecycle by making features consistent and versioned.

Delivered for a US property and home services technology groupHow we did it

Data drift caught before it reached production

MLflow, Great Expectations and Evidently wired into CI/CD to detect and block drift rather than discover it in a quarterly review.

Delivered for a US property and home services technology groupHow we did it

A pipeline that survives schema drift

BigQuery, Firestore and Feast aligned behind a contract-first design resilient to upstream changes outside our control.

Delivered for a US property and home services technology groupHow we did it

A production AI stack on GCP

Feature store, CI/CD, Kubernetes, IAM controls and observability — built to run, not to demonstrate.

Delivered for a US property and home services technology groupHow we did it

Explainable, business-aware models by construction

Connecting raw data, mathematical scoring and model learning so decisions can be defended, not just produced.

Delivered for a US property and home services technology groupHow we did it

A comparison experience users actually trusted

Double Diamond process applied to a consumer review journey, balancing credibility against conversion.

Delivered for a US property and home services technology groupHow we did it

Private AI with zero cloud dependency

Auralis: local model inference, hybrid retrieval and message-level provenance running entirely on infrastructure the customer owns.

Prestanda product engineeringHow we did it

We work under CMMI Level 3 and ISO/IEC 27001:2022. Client names and engagement specifics are disclosed only with written permission, which is why the work below is described by what it did rather than who it was for. If you would like to talk to a reference, ask us and we will arrange it directly.