Brendan O'Donnell
I help product teams ship reliable APIs, scale distributed systems, and pay down operational risk. Some of my best work has been in the messy seams between code and infrastructure where things tend to break.
Currently taking on small engagements — 5–10 hours/week, evenings and weekends.
15+ years building backend systems, including configuration management and deployment automation at a mobile security startup and Rackspace, email infrastructure at Mailgun, and clinical decision support APIs in health-tech. Strongest in API design, performance tuning, distributed systems, observability, and Python/Go codebases that have outgrown their original assumptions.
Design and implement HTTP and async APIs in Python (Django, Flask, FastAPI, Celery) and Go. Greenfield services or rescuing ones that have grown unmaintainable.
Profile, benchmark, and fix the SQL queries, async workers, and serialization paths that are actually costing you. OpenTelemetry, Pyroscope, Prometheus, and Grafana rollouts.
Untangle async integrations, replace deprecated dependencies, modernize the test and lint toolchain, and add the observability that lets you change things without fear.
Diagnose and patch upstream libraries, work with maintainers to land fixes, and own internal forks or pre-release builds when timelines demand it.
Health-tech SaaS · 2024–2025
A document processing pipeline was hitting a throughput ceiling weeks before a major customer rollout. I refactored it to support multiple simultaneous worker processes and fixed a poorly performing SQL query in the job loop. Then I built a JMeter benchmarking harness so the team could validate future changes without surprises. The rollout happened on schedule.
Health-tech SaaS · 2023–2024
Tesseract alone wasn't accurate enough on the patient reports feeding our NLP product. I built a hybrid pipeline combining Tesseract, SymSpell post-correction, and a TrOCR transformer model fine-tuned on a curated set of in-domain reports, and a benchmarking harness based on standard ISRI OCR evaluation tools. Result: 20% improvement in non-stop-word accuracy over Tesseract alone.
Email infrastructure at scale · 2017–2022
A core API for an email platform handling 1.5B+ messages/day was hitting scale ceilings on its original Python Flask implementation. I led a re-implementation in Go that handled significantly higher load on fewer resources, with a migration path that didn't require downtime or breaking existing clients.
Email infrastructure at scale · 2018–2020
Co-led a rewrite of the polling algorithm behind a high-volume sending pipeline. The new design was simpler, more predictable, and significantly more performant under load.
Multiple PRs merged into biocommons/hgvs, the reference Python library for HGVS variant nomenclature in clinical genomics:
Earlier OSS work includes contributions to gophercloud, mailgun/flanker, mailgun/holster, and several Chef cookbooks.
The best way to reach me is email. I respond within a business day.