Parallel Ant Colony Optimization (ACO) algorithm for the Travelling Salesman Problem (TSP), implemented fully in CUDA.
The project explores GPU parallelism, shared memory optimization, and data-parallel algorithms inspired by Cecilia & García (2012).
- 🚀 Implemented both Worker Ant and Queen Ant variants, leveraging shared memory and prefix-sum (Blelloch) parallel reduction for improved efficiency.
- ⚙️ Used atomic pheromone updates, CUDA Graphs, and curand for reproducible stochastic behavior.
- 📈 Achieved up to 3–5× faster execution compared to the Worker baseline, with stable numerical results on large datasets (
rat783,pr1002). - 🧪 Benchmarked on an NVIDIA Titan V GPU cluster with datasets up to 1000 cities.
Technologies: CUDA, C++, GPU Programming, Parallel Reduction, Shared Memory Optimization
Paper: Enhancing Data Parallelism for Ant Colony Optimisation on GPUs (Cecilia & García, 2012)
Implemented a distributed Δ-Stepping algorithm for the Single Source Shortest Paths (SSSP) problem using MPI.
The project demonstrates scalability trade-offs in distributed-memory graph processing.
- 🧮 Designed an efficient message-passing scheme with
MPI_Alltoallvfor inter-process bucket synchronization. - ⚙️ Integrated hybridization and edge classification heuristics, yielding up to 20–50% performance gain over the baseline.
- 📊 Conducted Δ-parameter tuning and weak scaling benchmarks up to 80 MPI workers, analyzing communication overhead and workload balance.
- 🌐 Generated test graphs via RMAT-1 and RMAT-2 models, following Chakaravarthy et al., “Scalable SSSP Algorithms for Massively Parallel Systems” (IPDPS 2014).
Technologies: C++, MPI, Parallel Computing, High-Performance Benchmarking
Paper: Scalable Single Source Shortest Path Algorithms for Massively Parallel Systems (Chakaravarthy et al., 2014)
Designed and implemented a distributed MapReduce system deployed on Google Cloud Kubernetes Engine (GKE).
The project demonstrates the full workflow of scalable data processing — from local binaries to containerized orchestration.
- ⚙️ Implemented Map, Partitioner, Combiner, and Reduce stages as statically linked binaries, ensuring portability and performance.
- 🐳 Containerized each component using Docker, and deployed them via Kubernetes manifests on GKE.
- ☁️ Automated deployment using Google Artifact Registry and shell scripts (
build_docker.sh,deploy.sh) for reproducible builds. - 🧩 Developed a Java client for orchestrating and monitoring distributed tasks through network communication with the master node.
- 🧠 Explored distributed file partitioning, intermediate combining, and fault-tolerant reduce operations at cluster scale.
Technologies: Java, C++, Docker, Kubernetes (GKE), Google Cloud Platform, Bash Automation
Interactive Weather Map Web App developed as part of the Web Applications course at MIMUW.
Visualizes real-time weather and geospatial data using modern, reactive frontend architecture.
- 🌦️ Integrated Overpass and Weather APIs for dynamic map-based weather visualization.
- 🔁 Utilized Redux and Redux-Observable for state management and asynchronous data flow.
- 💨 Built a responsive, TailwindCSS-based UI focused on clarity and performance.
- ⚡ Designed with modular architecture and component reusability for scalability.
Technologies: React, TypeScript, Redux, Redux-Observable, TailwindCSS, API Integration
