I’m Simon Bracegirdle — a software engineer in Perth, Western Australia. I’ve spent 20+ years building products: education software at SEQTA, cloud consulting at Mechanical Rock, and now government-relations tooling at GovConnex.
AI engineering
Since 2024 most of my work has been AI features that ship to paying customers. The biggest is a research assistant that orchestrates agents and tools, with a deep-research mode for longer questions. I also built the automated briefings that read the day’s political events and write a summary overnight, and the embeddings pipeline behind semantic search.
I’ve built the parts around them too: per-user cost caps and usage reporting, evals to catch regressions, prompt caching to hold down latency and spend, and hardening against prompt injection.
LLM agents tool use embeddings evals prompt design
Web applications
React and TypeScript on the front end, Node with NestJS and GraphQL behind it, and MySQL and Elasticsearch for storage and search. I’ve shipped a Microsoft 365 email integration for our CRM, and realtime alerting built on Elasticsearch percolation queries. I also look after the newsletter and monitoring pipelines, and the scrapers that collect parliamentary data.
TypeScript React NestJS GraphQL Elasticsearch MySQL
Cloud and observability
I work in AWS most days: ECS services, Lambda and SAM stacks, SQS pipelines, Cognito for auth. In 2022 I rolled out OpenTelemetry across the platform, including the tail-based sampling that brought our trace ingestion bill down. I want systems that tell you what’s wrong while there’s still time to fix it. I’ve also written up my take on OpenTelemetry.
AWS serverless OpenTelemetry Datadog CI/CD
Migrations
A lot of my better work has been deletions. I moved a production webapp from JavaScript to TypeScript without pausing feature work. Since then I’ve taken the e2e suite from Cypress to Playwright, swapped ad-hoc SCSS for design tokens, and removed legacy features once their replacements proved out. I’ve found this kind of work goes better done continuously than saved up for a scheduled project.
Team tooling
I spend a fair bit of time on the things that make a team faster: CI/CD pipelines, load testing, parallel test jobs, and code review that goes past LGTM. Lately that’s meant agent-assisted workflows too — review agents, definition-of-done gates, and docs written so both people and machines can follow them. The habit under all of it is to ship the thing.
Selected work
The last five years, most recent first:
- 2026Agent workflows, deep-research mode, and prompt-injection hardening
- 2025AI research assistant from prototype to production
Report generation, usage caps and admin reporting, and a vector-search service.
- 2024Webapp to TypeScript, and the first embeddings work
The migration ran to about 30,000 lines.
- 2023Realtime alerting and CRM email integration
Elasticsearch percolation for instant matching, plus search-term tokenisation.
- 2022OpenTelemetry rollout — collector sidecars and tail-based sampling
Elsewhere
My code is on GitHub and the rest of my writing is on the blog. The home page has what I’m reading.