~/experience/bmt.md

Software Engineer (DevOps Lead)

Balanced Media & Technology

Nov 2020 – Aug 2026 · 5y10m · full-time

Ran DevOps and platform engineering for the Hewmen AI data platform: EKS, CI, IAM and developer environments.

## highlights

• ▸ Cut CI builds for the main platform monorepo from 40 to 8 minutes using Docker layer caching and sccache for Rust, without carrying stale artifacts between runs; production deploys stopped taking all day.
• ▸ Software engineer and DevOps lead (senior of two, training the junior engineer into the role) on the Hewmen AI data platform: 11 services on PostgreSQL, MySQL and DynamoDB, with CI and clusters used by 12 engineers across departments.
• ▸ Operated the production and two staging Amazon EKS clusters, plus per-developer environment clusters, all defined in CDK for Terraform in TypeScript (Helm and Kubernetes providers) and deployed through CircleCI with manual approval gates.
• ▸ Owned AWS IAM for the whole company, writing every policy and role to least privilege.
• ▸ Built and ran the infrastructure for Hewmen, an AI platform-as-a-service on the OpenAI API that lets businesses chat securely with their own sensitive company data: a broker/worker API on EKS, customer-uploaded documents, and connectors into on-premises data sources.
• ▸ Kept customer data inside its boundary: on-prem sources stayed on customer hardware behind the connector, and the cloud side ran on EKS with least-privilege IAM written for every service and role.
• ▸ Connected customer on-premises data sources to the platform's broker/worker API on EKS, writing the Java connector and co-designing its TCP protocol, so OpenAI-backed LLM agents could work over private data on-prem and in uploaded documents.
• ▸ Operated the AI platform in production: CDK for Terraform provisioning, CircleCI deploys with approval gates, Prometheus and Grafana monitoring, on-call and written postmortems.
• ▸ Built Aegis, the team's development environment app, in Rust (Tauri, with a Svelte UI) and Docker; its container was later reused as the CircleCI runner so local and CI builds matched.
• ▸ Moved Aegis to a declarative NixOS flake with pinned packages, scripts and helper tools written in Rust: onboarding on company hardware went from several days of fixing issues to nearly instant.
• ▸ Introduced Prometheus and Grafana monitoring to the platform and used it on-call for incident detection, alongside performance analysis and AWS cost-anomaly detection.
• ▸ Introduced Prometheus and Grafana monitoring to the platform; carried on-call for production as the lead of two and wrote every postmortem, using the same monitoring for performance analysis and AWS cost-anomaly detection.
• ▸ Carried on-call for the production platform as the lead of two DevOps engineers and wrote every postmortem.
• ▸ Replaced manual delivery with a standard CircleCI pipeline across the organization's services.
• ▸ Wrote Rust libraries for the main platform monorepo and contributed to several of its services.
• ▸ Wrote a headless OBS system in Go and Docker on EC2, driven by a Twitch chat bot that starts streams and switches scenes, so anyone could demo Twitch extensions live without installing OBS or handling credentials.
• ▸ Built Sponster in React: a Twitch extension with character-driven animated overlay messaging, a message construction tool with scheduling, and a web dashboard streamers use to control it.