Built decode fast paths, fused attention-output RoPE quantization, device-side greedy selection, and synchronization reductions. Benchmarked 66–70× faster prefill and 115–133× faster decode over the fork baseline while preserving deterministic greedy output.
Stephen Murray
VP Engineering & Cofounder — principal-level systems builder
Boston, Massachusetts github.com/spmurrayzzz linkedin.com/in/stephenpmurrayProfile
Software engineer with 20+ years of experience building distributed systems. Cofounded Starry and led its engineering organization while working across cloud services, network infrastructure, embedded devices, customer products, developer tooling, and applied ML.
20+Years building production systems
50+Engineers led and mentored
99.995%Average platform availability
~2,000Production servers operated
Experience
VP Engineering & Cofounder · Starry Internet
2014–Present · BostonNationwide fixed-wireless internet service provider · Acquired by Verizon, Q1 2026
- Platform architecture. Built the service-oriented software platform spanning AWS and on-prem infrastructure, network control, embedded devices, customer systems, and operational tooling.
- Applied ML. Led network fault-localization, anomaly-detection, time-series forecasting, and churn-prediction work that reduced manual triage and improved QoS analysis.
- Reliability. Sustained 99.995% availability across roughly 2,000 servers through SLOs, incident response, postmortems, observability, and disciplined releases.
- Operations tooling. Shipped LLM-driven outage analysis and operations Q&A agents with incident, customer, log, and metrics tools plus Langfuse tracing.
- Network automation. Designed and contributed to IPAM/DCIM, RADIUS address allocation, TR-069 device management, SNMP automation, monitoring, and alarming.
- Service foundations. Helped grow 150+ Node.js services and roughly 5M lines of code, with shared foundations for auth, data, CI/CD, logging, and deployment.
- Team leadership. Hired and managed 50+ engineers, tech leads, and managers while mentoring senior ICs and reviewing architecture, code, and production changes.
- Product launches. Built vendor relationships across hardware and software and supported new-product introduction from early prototypes through nationwide operation.
Sustained hands-on contributionGitHub activity across Starry’s private engineering organization since 2014.
10,000+Commits
1,700+PRs authored
2,000+PR reviews
Recent Public Work
Designed and built a local workspace for concurrent coding-agent sessions with live runtime reconciliation, model controls, transcript editing and forking, subagent delegation, persistent memory, an embedded terminal, and HTML export.
Built a library that turns Git history into fill-in-the-middle training examples. Supports PSM, SPM, ZED, and mixed formats with AST-aware cursor placement, quality filters, synthetic negative examples, and a dataset browser.
Built a snapshot engine that stores worktree checkpoints as Git tree objects in a separate object database, with diff, selective restore, aliases, cleanup, and asynchronous Node.js bindings.
Built a project-scoped task system for AI agents with dependency cycle detection, Git reconciliation, atomic snapshots, bulk operations, tags, archiving, and SQLite persistence.
Earlier Experience
2012–2014
Lead Software Engineer · Aereo
Services for live OTA television streaming, “10-foot” Smart TV products, release tooling, CI/CD, and antenna and transcoder integrations.
2011–2012
Lead Software Engineer · DailyFeats
Full-stack product engineering for a consumer behavior and rewards platform.
2009–2011
Lead Engineer · Simply Interactive
Web products and technical leadership for custom in-house CMS. Used by Amazon, Asurion, and other clients.
2004–2009
Contract Software Engineer
Product, infrastructure, and web systems for a range of clients.
2002–2006
Professional Musician · Touring Drummer
Performed and toured professionally while beginning a software career.
Capabilities
AI engineering
Agent architecture, tools, evals, trace analysis, local inference, CUDA optimization, context engineering, FIM and fine-tuning datasets.
Platforms & reliability
AWS, Linux, Docker, CI/CD, service architecture, event systems, observability, SLOs, incident response, and release engineering.
Networks & devices
IPAM/DCIM, RADIUS, TR-069, SNMP, millimeter-wave RF, Wi-Fi, embedded Linux, hardware integration, and fleet operations.
Languages & data
JavaScript/TypeScript, Python, Rust, Go, C/C++/CUDA, SQL, MongoDB, Redis, Postgres, and MySQL.