Selected work

I designed and led development of an internal agentic platform that uses LLMs to automate digital marketing workflows at scale — work that previously required substantial manual operator time.

The engineering challenge was less about model choice and more about making non-deterministic components behave acceptably inside a deterministic product: bounding agent autonomy, designing for partial failure, keeping token cost predictable, and building enough observability that a failed run can actually be diagnosed rather than guessed at.

Built on cloud-native services across AWS and GCP with scalable microservices.

A platform I built to automate validation across diverse document types. It pairs a file repository and document-extraction service with integrations to the major cloud storage providers, then runs high-throughput background analysis using LLM reasoning.

Validation rules are fully configurable through the UI rather than hard-coded, and dashboards give real-time visibility into validation runs, throughput, and outcomes. The design goal was to keep the LLM confined to the judgment step while everything around it — queueing, retries, auditability — stayed conventional and testable.

Happy to walk through the architecture or demo it on a call.

I led an international IBM team building a custom solution for Bell, the Canadian telecom operator: integrating existing and new Java applications with Kafka queues and an ELK stack to add business activity monitoring across the estate.

Scope covered Logstash pipeline design, Kibana dashboards, alerting, and machine learning jobs for anomaly detection. Two successful milestone deliveries to production.

This is the work that most directly informs how I approach LLM observability today.

Eight years across three roles. As Senior Engineer and then Solutions Architect I designed business systems architecture and led development teams through major upgrades of national JIS and OSS platforms, including technical proposals and public tender bids.

As Engineering Manager for ICT Solutions I ran a department of 25 IT professionals: scaled the team from 15 to 25, delivered four national projects on time and within budget, and held a 90% employee satisfaction index.

Long-horizon systems with real consequences for failure — a useful counterweight to the move-fast instincts of AI tooling.