- Architected and operated a 16TB (8TB usable) NAS on TrueNAS Scale sustaining up to 1 Gb/s LAN reads/writes.
- Built a RAID Z2 ZFS storage pool with automated snapshots and cloud backups to protect data integrity.
- Hosted 15+ containerized services via Docker Compose with Grafana and NetData monitoring.
I am a 20-year-old self-taught Computer Science student at Vrije Universiteit Amsterdam (GPA: 9.5 / 10), specializing in machine learning research, distributed systems, and high-performance (scientific) computing. Alongside my BSc, I take extra-curricular MSc coursework in Parallel Programming and Distributed Systems.
Previously, I worked as an IT Operations Intern at SURF building multi-node Proxmox/Ceph clusters and local LLM inference testbeds, and received an NLnet Foundation grant for redundant eduVPN infrastructure. You can find my code on GitHub, connect on LinkedIn, or view my CV (PDF).
Projects
- Built an agentic evaluation pipeline in LangGraph to benchmark LLMs against HumanEval via a ReAct-style repair and evaluation loop.
- Implemented sandboxed Docker execution, streaming JSONL results, and multi-iteration repair runs for deterministic pass@1 measurement.
- Built an interactive 60 FPS trading terminal in Processing 4 backed by a live Coinbase WebSocket feed.
- Engineered an append-only Structure-of-Arrays double-entry ledger with O(1) time-travel and revert synced to the market tape.
- Split hot-path scans onto Java 21 Vector API (SIMD) kernels and quantitative risk metrics (Volatility, 95% VaR, Sharpe Ratio, Max Drawdown) onto Apache Spark SQL.
- Implemented a parallel chunk-based HTTP Range downloader supporting HTTP/2 (
h2/h2c) with multi-threaded and multi-process JVM chunk writers. - Added retry logic, KoTest coverage, and a powers-of-two thread benchmarking CLI with plotting scripts.
- Developed a parallel Mandelbrot renderer using Julia's
Distributedcomputing framework with a master-worker architecture and dynamic load balancing viaRemoteChannels. - Achieved up to 8.3x speedup over the sequential baseline with sub-150ms parallel overhead on the DAS-5 supercomputer running SLURM.
- Implemented a bare-metal virtual machine interpreter in C to execute compiled IJVM bytecode binaries.
- Engineered the fetch-decode-execute cycle, local variable frames, operand stack, dynamic memory layout, and strict bounds checking against buffer overruns.
- Created deployment scripts and Proxmox NixOS VM templates using Hydra VMA images for SURF infrastructure.
- Tested and patched across personal and SURF clusters; partially funded by the NLnet Foundation.
- Developed modular plugins (Cogs) extending Red-DiscordBot with custom integrations and async tooling.
- Built an interactive Python tool for tracing curves, fitting Bezier curves, and visualizing Bernstein polynomials.
Research & Reports
HTTP Flood DoS attack detection model trained in TensorFlow/Keras with a NumPy, Pandas, and Matplotlib preprocessing pipeline; achieved 95.4% accuracy on test data and 92.2% on an unknown dataset.
Performance study of dynamic load balancing with Julia's Distributed framework
and RemoteChannels on the DAS-5 supercomputer.
Experimental physics study investigating LED luminance vs. voltage with Kolmogorov-Smirnov statistical validation and Python/Matplotlib analysis.
6th-order Bezier curve models for rollercoaster safety and physiological analysis using derivative-based velocity and acceleration computation in NumPy/SciPy.
Experience & Grants
- Architected and deployed a multi-node Proxmox VE High Availability compute cluster backed by a distributed Ceph storage pool for research workloads.
- Led zero-data-loss major-version hypervisor migrations from Proxmox v8 to v9 with dedicated Proxmox Backup Server snapshotting and disaster recovery.
- Provisioned and benchmarked local LLM inference pipelines (Qwen 3.8, DeepSeek v4 Flash) as an isolated testbed for adversarial AI security research and automated red-teaming.
- Executed physical migrations, rack management, and structured cabling across Digital Realty, Nikhef, Tilburg University, and Radboud University.
- Awarded independent grant funding from the NLnet Foundation to architect an HA Proxmox compute cluster hardening core eduVPN infrastructure against single points of failure.
- Executed zero-downtime v8 to v9 hypervisor migration with distributed snapshotting and benchmarked fault tolerance, I/O throughput, and sub-millisecond network latency.
- Contributed to university infrastructure services; fixed critical bugs in
eat-apiand added Heilbronn campus support atmenu.tum.sexy. - Collaborated with a team of 9 developers via code reviews, issue tracking, and process improvements.
- Delivered software for 30+ clients, engineering asynchronous event-driven Python modules handling websocket events with sub-100ms latency.
Education
Technical Skills
- Languages: Python, Julia, C/C++, AT&T x86-64 Assembly, Scala, Kotlin, MATLAB, JavaScript, Bash
- Systems / Ops: Linux, Proxmox VE, Proxmox Backup Server, Ceph, TrueNAS (ZFS), Docker (Compose), SLURM, Git, CI/CD (GitHub Actions), Prometheus, Grafana, NetData, OCI
- Frameworks / Libraries: PyTorch, TensorFlow/Keras, Apache Spark, LangGraph, MPI/OpenMP, CUDA, NumPy, Pandas, SciPy, Matplotlib, OpenCV, Django, Flask