Software Project Showcase

A distributed compute orchestration platform for machine learning workloads. Registers and monitors GPU/CPU compute nodes, schedules job submissions against priority, fair-share and backfill algorithms, and tracks workloads through their full lifecycle. Four role-specific dashboards - System Administrator, ML Researcher, Resource Manager and Operations Monitor - sit on a JWT-authenticated REST API, backed by Delta Prism, a UV-built property graph database. Built in a one-week push in June 2025 and untouched since; currently dormant, not under active development.

• Compute-node registration and heartbeat monitoring, with pluggable FIFO, priority, fair-share and backfill scheduling algorithms
• Full workload lifecycle - submit, start, cancel, complete - with per-job CPU, GPU, memory and duration requirements
• Four role-specific dashboards (System Administrator, ML Researcher, Resource Manager, Operations Monitor) on one authenticated API
• JWT authentication with role-based access control across admin, operator and user roles

and 1 more

A machine learning benchmark project demonstrating optimized ML performance across different hardware configurations. Showcases efficient GPU utilization for medical image scanning with modest compute resources.

• Performance comparison across multiple GPU types
• Resource utilization optimization
• Efficient ML model deployment

A public website and hosting infrastructure for Akademia Mistrzow, a Gdynia-based youth sports foundation - GDPR-compliant contact form with AWS SES email notifications, Google Analytics and Search Console integration, and a 100/100 Lighthouse score across every category.

• GDPR-compliant contact form with 6-month database retention and AWS SES email notifications
• Dedicated Hetzner VM (Apache, PHP, MariaDB) with SSH key-only access and a UFW firewall
• Google Search Console and Analytics integration across two domains
• 100/100 Lighthouse score across Performance, Accessibility, Best Practices, and SEO

A hybrid deployment of large language models across both private cloud immersion clusters and AWS. Features fine-tuned Meta Llama 3 models with consistent performance across environments.

• Cross-platform LLM deployment
• Fine-tuned model training
• Consistent performance across environments

A self-hosted language model inference platform running on private GPU infrastructure, giving internal products model access without sending data to external providers. Four GPU nodes serve different models behind a single OpenAI-compatible endpoint, with load balancing, per-developer session management and rate limiting handled by a dispatcher layer.

• Single OpenAI-compatible API endpoint fronting four independently served models
• Request routing with least-load, round-robin and manual balancing strategies
• Per-developer session management with idle timeout and rate limiting
• Administrative web console for nodes, sessions, developers and audit history

and 2 more

An internal engineering portal that consolidates technical documentation and operational tooling behind a single interface. The Forge pairs a searchable document library with a catalogue of infrastructure management tools, and exposes both through a token-authenticated REST API so other systems can query them programmatically rather than through the browser.

• Searchable document library with in-browser viewing and direct download
• Catalogue of infrastructure and cloud management tools with per-tool metadata
• Token-authenticated REST API for programmatic access to documents and tooling
• Sandboxed file handling with path validation and access controls

and 1 more