Course Project – Theories & Methods of Computational Intelligence (Lviv Polytechnic, 2025)

1. User Spending Prediction – Engineered and benchmarked Linear/Logistic Regression, Decision Tree, Random Forest, Gradient Boosting using pandas, NumPy, scikit-learn, Matplotlib with R?, MAE, MSE, Accuracy.

2. Opossum Clustering Analysis – Reduced features with PCA and clustered via K-Means, DBSCAN, Fuzzy C-Means using pandas, NumPy, scikit-learn, Matplotlib, Seaborn with Silhouette, Davies–Bouldin, Calinski–Harabasz.

3. Name Matching with Fuzzy Logic – Built fuzzy logic name-matching pipeline using fuzzywuzzy, Soundex, Damerau–Levenshtein, scikit-fuzzy in Python with composite similarity scoring.

4. Heart Disease Prediction – Trained single-layer Perceptron with Sigmoid using PyTorch, NumPy, StandardScaler with Accuracy and Loss evaluation.

5. Image Classification with CNN – Built custom CNN in PyTorch with manual conv/pooling, Adam, CrossEntropyLoss with Accuracy and Loss tracking and visualization.

6. Neural Networks for NLP – Generated BERT embeddings via Hugging Face, processed Ukrainian text with spaCy, compared semantic similarity using cosine similarity and GloVe.

----------------------------------------------------------------------------------------------------
Kaggle Project – House Prices Prediction (2025)

Performed EDA, imputed missing values, removed outliers, engineered features (TotalSF, Age), one-hot encoded categories, log-transformed target, trained tuned XGBoost Regressor with RMSLE validation.

----------------------------------------------------------------------------------------------------

Course Project – Neural network technologies & systems (Lviv Polytechnic, 2025)

1. Transformer Text Generation Model - Developed a PyTorch-based Transformer Decoder with BPE tokenization and custom positional encoding, optimizing hyperparameters for efficient text prediction.

2. Local LLM Integration - Developed a Python CLI for local Gemma inference via LM Studio and OpenAI-compatible API, implementing dynamic parameter tuning.

3.RAG Knowledge System - Engineered a LangChain pipeline using ChromaDB and local LLMs via Ngrok, utilizing MMR for precise document retrieval.

4. Generative AI (Img2Img) - Implemented a Stable Diffusion v1.5 Image-to-Image pipeline using PyTorch and Diffusers for automated visual style transfer and transformation.
5. Multimodal Vision Stack - Python, PyTorch, Transformers, Accelerate, Safetensors, Pillow, CLIP, and LLaVA-1.5 model utilized in Google Colab to process images and generate text descriptions.
6. Multi-Agent System Design - Python, CrewAI, LangChain and OpenAI client orchestrate Researcher, Analyst and Writer agents to generate structured technical reports using local large language models.

7. Generative Art-Critic Agents - Python, CrewAI, ngrok, Stable Diffusion, and LLaVA empower autonomous agents to create and critique images using local large language model inference.