Wiktoria
Przykucka
creative technologist
Wiktoria
// aboutAbout WiktoriaReal-world AI
// about
About Wiktoria
Creative Technology & AI student building end-to-end intelligent systems — from data pipelines and models to deployment on edge devices and in the cloud.
I work across software, hardware and applied ML: Python, RAG, computer vision, microservices with Docker and Kubernetes, and edge AI on Raspberry Pi.
- Software + hardware + intelligent systems
- Python · RAG · computer vision
- Edge AI on Raspberry Pi
- Microservices · Docker · Kubernetes
- End-to-end ML: data → model → deploy
// goalAspiring Scientist
// goal
Aspiring Scientist
Long-term goal: contribute to research at the intersection of AI, systems and real-world sensing — rigorous experiments, reproducible pipelines, and work that holds up outside the lab.
// goalTop CS / Quantum programs
// goal
Top CS / Quantum programs
Targeting competitive computer science and quantum computing master's tracks — strong foundations in maths, physics and programming, with research-oriented project work.
// interestHorse AI — posture & lameness detection
// interest
Horse AI
Personal research interest: computer vision for equine posture and early lameness cues — combining biomechanics with pose estimation and time-series models on real riding footage.
// interestRobotics & quantum computing
// interest
Robotics & quantum computing
Exploring how embodied agents and quantum algorithms can extend classical ML — from Unity RL simulators to reading up on qubit-based optimisation and hybrid classical–quantum workflows.
// locationKortrijk, Belgium
// location
Kortrijk, Belgium
Currently based in Kortrijk for my BSc at Howest — internships and industry projects in Belgium.
// homeKoszalin, Poland
// home
Koszalin, Poland
Home city in Poland — secondary school (extended maths & physics) and family base between semesters.
// statusOpen to internships · AI / ML roles
// status
Open to opportunities
Available for internships and student roles in AI, ML, computer vision and applied RAG — especially teams building real products, not just slide decks.
// languagePolish — Native
// language
Polish — Native
Native speaker — comfortable in technical and academic contexts, presentations, and bilingual project documentation.
// languageEnglish — Above-intermediate
// language
English — Above-intermediate
Studying and working in English daily — coursework, industry internships, technical writing and team communication.
// educationHowestBSc Creative Tech & AI
// education
Howest — BSc Creative Technology & AI
2024–2027 · Kortrijk, Belgium. Hands-on programme mixing AI/ML, software engineering, IoT and creative technology — project-led from day one.
- AI · ML · Deep Learning
- Databases · IoT · Networking
- Docker · microservices
// educationII LO BroniewskiMatura
// education
II LO im. Władysława Broniewskiego
2019–2024 · Koszalin, Poland. Secondary school with extended mathematics and physics — foundation for engineering and CS studies.
- Extended Maths & Physics
- Matura diploma
// experienceActemiumStudent AI / RAG Developer
// experience
Actemium — Student AI / RAG Developer
2026 · Belgium. Industry internship building production RAG systems from messy Excel exports to searchable, dockerized chat interfaces.
- Excel → searchable RAG pipelines
- Hybrid RAG · Neo4j knowledge graph
- PostgreSQL + pgvector
- Azure deployment · Dockerized services
// recognitionTech Connect expo — 2 projects
// recognition
Tech Connect expo
Presented two projects at the Tech Connect expo — ASL recognition on Raspberry Pi and an LLM-powered workflow — demoing live to visitors and industry guests.
// project
trAInerPose AI · LSTM
// project
trAIner
Gymnastics coaching assistant: extracts pose landmarks with MediaPipe, segments routines with an LSTM, and scores movement quality against reference grades.
Shipped as a dockerized stack with FastAPI backend and Gradio UI for coaches to upload clips and get structured feedback.
- MediaPipe + LSTM pose pipeline
- 91% segmentation accuracy
- 86% grade accuracy
- FastAPI · Gradio · Docker
// project
ASL RecognitionMediaPipe · edge
// project
ASL Recognition
Real-time American Sign Language letter recognition on a Raspberry Pi — MediaPipe hand landmarks fed into a lightweight classifier, rendered with OpenCV and served through Flask.
Built for edge deployment: low latency, no cloud dependency, demoed at Tech Connect.
- MediaPipe hands · 100% test accuracy
- Raspberry Pi edge inference
- OpenCV · Flask · real-time video
// project
RAG / GraphRAGHybrid RAG · NDA
// project
RAG / GraphRAG
Industry RAG system for Actemium (under NDA): ingests structured Excel exports, builds hybrid retrieval over PostgreSQL/pgvector and a Neo4j knowledge graph, and answers via an Azure-hosted chatbot.
End-to-end ownership from data cleaning and embedding pipelines to dockerized deployment and evaluation.
- Hybrid retrieval · GraphRAG
- Neo4j · PostgreSQL + pgvector
- Azure deployment · Docker
// project
Show Jumping RLUnity ML-Agents
// project
Show Jumping RL
Low-poly show-jumping simulator in Unity where an agent learns to navigate courses via ML-Agents — curriculum learning and reward shaping for stable gaits and clean jumps.
Custom C# environment logic, procedural obstacles, and training loops tuned for sample efficiency on consumer hardware.
- Unity ML-Agents
- Curriculum learning · reward shaping
- C# environment · low-poly horse asset
// langPython
// lang
Python
Primary language for ML, RAG pipelines, FastAPI/Flask backends, data wrangling and automation across coursework and industry projects.
// langSQL
// lang
SQL
Relational modelling, queries and migrations — PostgreSQL in production RAG stacks and normalized schema design (3NF).
// langC#
// lang
C#
Unity game logic and ML-Agents environments — gameplay systems, reward functions and simulation tooling.
// ai·mlMachine Learning
// ai·ml
Machine Learning
Supervised pipelines end-to-end: feature engineering, training, evaluation and deployment — scikit-learn, PyTorch and TensorFlow.
// ai·mlComputer Vision
// ai·ml
Computer Vision
Pose estimation, hand tracking, segmentation and real-time inference — MediaPipe, OpenCV, YOLO on desktop and edge devices.
// ai·mlRAG
// ai·ml
RAG
Retrieval-augmented generation: chunking, embeddings, vector search and orchestration for domain-specific chatbots.
// ai·mlGraphRAG
// ai·ml
GraphRAG
Hybrid retrieval combining knowledge graphs (Neo4j) with vector search for structured enterprise data.
// ai·mlReinforcement Learning
// ai·ml
Reinforcement Learning
Policy learning in simulated environments — Unity ML-Agents, reward design and curriculum schedules.
// ai·mlPCA
// ai·ml
PCA
Dimensionality reduction for visualization and feature compression before downstream models.
// ai·mlPrompt Engineering
// ai·ml
Prompt Engineering
Structured prompts, tool use and evaluation for LLM-backed assistants in RAG workflows.
// ai·mlApplied & Gen. AI
// ai·ml
Applied & Generative AI
Production LLM workflows — chatbots, tool use, structured outputs and domain-specific assistants beyond basic prompting.
// ai·mlGuardrails
// ai·ml
Guardrails
Input/output validation, content filters and safety layers for LLM apps — used in industry RAG chatbot delivery.
// ai·mlEdge AI
// ai·ml
Edge AI
Training and deploying compact custom models on Raspberry Pi and constrained hardware for real-time inference.
// ai·mlMLOps
// ai·ml
MLOps
Azure ML pipelines, automated cloud training jobs, CI/CD for models, and Hugging Face deployment workflows.
// frameworkPyTorch
// framework
PyTorch
Deep learning for sequence models and custom training loops — LSTM pose segmentation in trAIner.
// frameworkTensorFlow
// framework
TensorFlow
Neural network coursework and experimentation alongside PyTorch stacks.
// frameworkscikit-learn
// framework
scikit-learn
Classical ML baselines, preprocessing and model selection for tabular and feature-vector data.
// frameworkpandas
// framework
pandas
Data cleaning and feature tables — especially Excel ingestion for RAG pipelines.
// frameworkNumPy
// framework
NumPy
Numerical arrays underpinning CV feature vectors and ML preprocessing.
// frameworkOpenCV
// framework
OpenCV
Capture, draw and post-process video frames for real-time ASL and pose demos.
// frameworkMediaPipe
// framework
MediaPipe
Pose and hand landmark extraction for trAIner and ASL — fast enough for edge devices.
// frameworkYOLO
// framework
YOLO
Object detection experiments for coursework and rapid prototyping.
// frameworkFlask
// framework
Flask
Lightweight HTTP APIs and video streaming endpoints on Raspberry Pi.
// frameworkFastAPI
// framework
FastAPI
Typed async APIs behind trAIner inference and microservice boundaries.
// frameworkGradio
// framework
Gradio
Quick demo UIs for model inference — coaches upload video, get scores back.
// toolDocker
// tool
Docker
Containerized ML services and RAG stacks for reproducible deploys.
// toolDocker Compose
// tool
Docker Compose
Multi-service local stacks — API, vector DB and graph DB wired together.
// toolKubernetes
// tool
Kubernetes
Orchestration concepts for scaling microservices — coursework and cloud labs.
// toolGit
// tool
Git
Version control for team projects, feature branches and portfolio repos on GitHub.
// toolAzure AI
// tool
Azure AI
Cloud-hosted LLM and deployment targets for industry RAG chatbots.
// toolOllama
// tool
Ollama
Local LLM inference for prototyping RAG without cloud spend.
// toolPrivateGPT
// tool
PrivateGPT
Offline document Q&A experiments with local embeddings and models.
// toolJupyter
// tool
Jupyter
Exploratory notebooks for EDA, model tuning and coursework deliverables.
// toolCursor
// tool
Cursor
AI-assisted IDE for faster iteration on portfolio and project codebases.
// toolMCP
// tool
MCP
Model Context Protocol servers — connecting LLMs to external tools, APIs and data sources.
// toolROS
// tool
ROS
Robot Operating System — nodes, topics and integration for robotics coursework and perception pipelines.
// toolHugging Face
// tool
Hugging Face
Model hosting, Hub workflows and deployment for trained checkpoints as part of MLOps pipelines.
// toolCI/CD
// tool
CI/CD
Automated build and deploy pipelines on Azure — model training jobs, image builds and release workflows.
// dataPostgreSQL
// data
PostgreSQL
Primary relational store with pgvector extension for embedding search in RAG.
// dataNeo4j
// data
Neo4j
Knowledge graphs for GraphRAG — entities, relations and Cypher queries.
// dataInfluxDB
// data
InfluxDB
Time-series storage for IoT sensor coursework and monitoring dashboards.
// dataGrafana
// data
Grafana
Dashboards on top of InfluxDB and metrics pipelines.
// dataElastic Stack
// data
Elastic Stack
Search and log analytics — indexing and querying unstructured text.
// dataKibana
// data
Kibana
Visualization layer for Elasticsearch indices and operational dashboards.
// data3NF Design
// data
3NF Design
Normalized relational schemas — reducing redundancy before scaling data layers.
// hardwareRPi GPIO
// hardware
RPi GPIO
GPIO programming on Raspberry Pi for sensors, LEDs and IoT coursework.
// hardwareI²C
// hardware
I²C
Serial bus for attaching sensors and displays to microcontrollers and Pi boards.
// hardwareSensors
// hardware
Sensors
Reading environmental and motion sensors into IoT pipelines and dashboards.
// hardwareBLE UART
// hardware
BLE UART
Bluetooth Low Energy serial bridges for wireless sensor prototypes.
// hardwareRobotics
// hardware
Robotics
Robot software stacks with ROS — perception, control and CV pipelines on physical platforms.
// creativeBlender
// creative
Blender
3D modelling for assets and visual coursework — low-poly and textured meshes.
// creativeUnity
// creative
Unity
Game engine for the show-jumping RL simulator — scenes, physics and ML-Agents integration.