RESUME

Summary

Founding ML systems engineer with 5+ years of software engineering experience, including 4+ years building production GPU platforms and infrastructure. Set product direction and led technical direction across execution, routing, serving, storage, and control-plane architecture for a multi-region serverless GPU platform.

Experience

Cerebrium (YC W22): Founding Engineer

Jan 2022 to May 2026 · New York, NY · Hybrid

• Built and operated production execution infrastructure across Python runtimes and multi-region Kubernetes/EKS clusters, supporting 5,000+ cumulative unique applications, 50,000+ daily GPU runs, and a peak managed fleet of 2,000+ GPU nodes; built custom images with Packer, containerd/Nydus, NVIDIA CDI, and gVisor.
• Customized a Knative Serving 1.18 fork for GPU workloads, reducing stale-endpoint 5xx responses by roughly 90% and improving p95 cold-start time from roughly 5 seconds to under 1 second through lifecycle signaling, quarantine, and routing changes.
• Led the storage architecture shift from EBS to an S3-backed JuiceFS POSIX filesystem with node-level and network caches plus cross-region mirrors; reduced operating costs by 90% while sustaining roughly 500 MB/s to 1 GB/s read throughput.
• Defined a cross-team runtime contract by creating Cerebrium’s initial Python runtime and a Rust sidecar linked by a custom Unix-socket protocol, isolating logging, lifecycle events, and platform I/O from user code.
• Bootstrapped and evolved a Go request gateway with 5 ms p99 overhead across auth and validation, HTTP/WebSocket proxying, async jobs, cancellation, run tracking, and OpenTelemetry tracing.
• Led cross-team design and rollout of Turso-backed metadata and routing state across backend and gateway services, coordinating dual writes, replica-freshness gates, backfills, and rollback tooling.
• Designed a read-only GPU capacity planner using Python, HiGHS MIP/LP optimization, Kubernetes state, and AWS pricing; a 30-day replay for one regional GPU fleet modeled 7.6% lower cost with zero underprovisioned planning intervals.
• Built TypeScript/AWS Lambda services, React interfaces, HTMX/templ Knative dashboards, and Cerebrium’s initial Python CLI, supporting roughly 1,000 deployments/month and real-time observability for application metrics and logs.

Invictus Capital: Data Scientist

Sep 2019 to Jul 2020 · Cape Town, South Africa · On-site

• Built latency-sensitive, multi-exchange trading systems and Kubernetes deployment infrastructure for Bitcoin arbitrage and margin-lending strategies.

Education

Cornell University: Master of Engineering, Electrical and Computer Engineering

2020 to 2021 · Grade: 4.05/4.3 · Merit Scholar · AI Focus

• Ant Neuroevolution: built a PyTorch multi-agent simulation and evaluated neuroevolution and reinforcement learning for pheromone-based foraging.

University of Cape Town: BSc (Hons) Computer Science · BSc Computer Science & Game Development

2015 to 2019 · First Class Honours

Honours thesis: Implicative Bayesian Networks.

Skills

Languages

Go, Python (primary); TypeScript, Rust (production); Zig (recent); Elixir, C++ (prior)

Systems and infrastructure

Kubernetes, EKS, Knative, Terraform, Docker, Helm, Karpenter, AWS, S3, JuiceFS, containerd, NVIDIA Container Toolkit/CDI, gVisor

ML, data, and observability

PyTorch, PostgreSQL, Redis, RabbitMQ, DynamoDB, Turso, OpenTelemetry, Prometheus, Grafana

Selected project

surfsk8s: multi-cluster Kubernetes TUI with informer-backed state, CRD discovery, structured filtering, and bounded rendering.