Projects

A selection of systems and tools I've built, from experimental projects to production deployments.

Featured

ZET Uživo

Real-time ZET tram and bus tracker for Zagreb. Shows live vehicle locations on a map together with stops and line information.

The system ingests GTFS-Realtime transit feeds, processes vehicle position updates, and streams them to the frontend for live map updates.
Designed and operated as a self-deployed production system, including containerized services, domain configuration, and a publicly indexed deployment.
Vehicles are rendered on a Leaflet map with route overlays, stop markers, and arrival predictions, updating continuously as new data arrives.

Production Infrastructure

Maestro

AI workflow orchestrator for structured software development using configurable pipelines, conventions, and engineering phases.

Designed to make AI-assisted software development deterministic by replacing ad-hoc prompting with structured, repeatable workflows.
Projects define their own pipelines, conventions, and phase prompts, allowing teams to encode development practices instead of relying on individual prompting styles.
Coordinates AI reasoning with deterministic scripts to create a consistent, maintainable development process that can be shared across projects.

Developer Tooling

queue-consumer

Symfony Messenger worker for background job orchestration. Supports YAML-defined jobs, queue routing, and lifecycle callbacks for building reusable processing pipelines.

Designed as a fully generic worker where jobs are defined through configuration rather than implemented as dedicated code, allowing new pipelines to be added without modifying the consumer itself.
Multiple worker instances can process separate message streams, enabling controlled parallelism and workload isolation.
Originally developed to safely process concurrent LLM prompts as part of a larger RAG infrastructure.

Distributed Systems

RAGu

Experimental Retrieval-Augmented Generation system exploring document summarization, ANN search, reranking, and MMR-based context selection.

Experimental RAG infrastructure designed to explore modern retrieval pipelines, including document summarization, vector search (ANN), reranking, and maximal marginal relevance for context selection.
The project deliberately favors architectural breadth over depth, serving as a playground for experimenting with distributed system components and retrieval strategies.
Built as a companion project for studying software architecture, using concepts from architecture literature to iteratively design, test, and refine system structure.

AI Infrastructure

Mojave Online

Experimental multiplayer framework for Fallout: New Vegas, focusing on player synchronization and real-time state replication in a single-player engine.

Research project exploring how to retrofit multiplayer functionality into a game engine originally designed for single-player gameplay.
Implemented as a C++ NVSE plugin that synchronizes player state, animations, and actions across clients while working within the constraints of the Fallout: New Vegas scripting and engine systems.
The project currently supports basic player synchronization and serves as an experimental foundation for building cooperative multiplayer systems on top of the existing engine.

Reverse Engineering

Other Projects