Lee Goymer

← Home · Ctrl/Cmd+P prints a clean PDF

Lee Goymer

Systems generalist · Applied-AI engineer — Brisbane, Australia lmgoymer@gmail.com · linkedin.com/in/lee-goymer · leegoymer.com · github.com/LeeGoymer

Profile

Computer Science graduate and ex-Amazon engineer who spent 2025–26 on a self-directed build year: designed, built, and operated a ~20-service AI platform solo on personal hardware — model-evaluation lab (8,351 models tracked, IRT-based test design, calibrated LLM judging), GPU scheduler with reversible preemption, verified knowledge-compression archive, multimodal edge assistant — all in production with monitoring, layered backups, and automated recovery. Strongest in ambiguous, end-to-end problems: architecture through implementation through honest measurement through operations.

Experience

Independent R&D — Personal AI Platform · 2025 – present, Brisbane

  • Built and operate ~20 production services on self-administered infrastructure (Ubuntu / systemd / FastAPI / PostgreSQL / nginx; RTX 3090 + ARM edge node): 3-tier backups to off-site object storage, deadman alerting, health-graph control plane.
  • Built an open-model evaluation lab: 8,351 models tracked, 246K benchmark results across 6 modalities; designed text benchmarks with 2PL Item Response Theory item selection; calibrated a local 31B LLM judge over 13.6K free-text answers; exposed and corrected an in-sample correlation inflation (ρ 0.935 → honest 0.607 → 0.682 after regrade).
  • Built a GPU lease scheduler with priority aging and reversible preemption (two-phase, VRAM-verified pause; RAM park ledger; tree-aware process enforcement) — arbitrates 3+ competing AI workloads on one GPU, failure-injection tested.
  • Built ARK: provenance-first knowledge-compression archive over open corpora (Wikipedia, Gutenberg, OpenAlex…), 15 validated scale rungs, 112/112 archives verified by full round-trip hashing; FTS / DuckDB / SPARQL query surfaces.
  • Shipped public web products end-to-end (OAuth2, WebSockets, 800+ req/min third-party API ingestion, TLS deployment) — e.g. completion.trainjumper.com.

Cloud Network Engineer — PCCW Global · 2024 · short engagement, Brisbane

  • Carrier cloud-network configuration for a global telecommunications provider; chose to exit during the probationary period over role fit.

Career note (2023 – 2025): took a deliberate break after Amazon, navigated a difficult market, and worked short engagements — including disability-support care work — before committing fully to the build year above. Ask me about it; it's a straightforward story.

Software Development Engineer — Amazon (Amazon Fresh) · Jul 2022 – Mar 2023, Brisbane (US grocery org)

  • Built and operated services for proactive customer-order-issue remediation across the US Amazon Fresh network — ingesting, processing, triaging, and routing failed-delivery, late-delivery, and order-issue events.
  • Contributed to the long-horizon migration of the legacy remediation system onto a newer internal platform.
  • Away-team work in a very large legacy Java codebase: updated a set of sensitive refund-eligibility edge cases, enabling legitimate customer refunds that were previously auto-rejected.
  • Contributed to the planning stage of a new freight-logistics system with global scope (away-team); standard SDE loop throughout — on-call rotation, system-health improvements, code reviews, intern onboarding and mentoring.

Dealer operations & media — automotive retail · 2025 – present (part-time/day job)

  • Photography, inventory data pipelines, and listing operations; built scraping/automation tooling for dealer-platform inventory data (resumable, deduplicating, image-pipeline aware).

Software Developer (placement) — RightCrowd · 2019, Gold Coast

  • Led development of a security web API framework and companion mobile app on Azure / .NET / Python / SQL as a university industry placement.

Digital Security Research Assistant — Griffith University IIIS · 2017 – 2019

  • Formal verification research on Ethereum/Solidity smart contracts (CSP#, PAT); security verification of a hypothetical trading platform.
  • Selected after top national first-year-team result in the Australian Cyber Security Challenge (CySCA); fintech penetration-testing engagement in Hong Kong with Entersoft (Hackfest 2018).

Education

B.Sc Computer Science — Griffith University · Graduated Feb 2020 · GPA 5.8/7 International Baccalaureate — Queensland Academy SMT · ATAR equiv. 94.9

Selected skills

AI/ML engineering: LLM evaluation & benchmarking (IRT, judge calibration, contamination policy) · local inference (llama.cpp, vLLM-class serving, GGUF/ONNX/RKNN quantization) · embeddings & retrieval · vision (SAM 2.1, depth models) · ASR/TTS pipelines · agentic/tool-use systems. Systems & platform: Python (deep), FastAPI, PostgreSQL, SQLite, Redis, systemd, nginx, Linux administration, restic/backup architecture, observability (Grafana, health state machines), GPU resource management (NVML). Security background: CTF (CySCA national top first-year team), formal verification (CSP#/PAT), web/app penetration testing. Also: C/C++, Rust, Java/JavaScript (Next.js), Go, embedded (Arduino, ARM SBCs).

References available on request. This resume prints cleanly — Ctrl/Cmd+P.