Research Assistant, Document Intelligence Systems, Czestochowa University of Technology
Duration: 2025 - 2026
Summary:
- Designed and evaluated large language model-based document understanding systems for extracting structured information from complex, layout-rich PDF documents
- Developed retrieval-augmented pipelines integrating transformer models, OCR, vision-language models, embeddings, and vector search for robust document question answering
- This project focused on improving document AI reliability and interpretability through advanced AI techniques
Responsibilities:
- Designed and evaluated document understanding systems using LLMs and OCR technologies.
- Developed retrieval-augmented generation pipelines combining multiple AI models and vector search.
- Implemented reproducible experiments with various performance and reliability metrics.
- Analyzed hallucination patterns and grounding failures to enhance system robustness.
Technologies: Large Language Models, Transformers, OCR (Tesseract, PaddleOCR), Vision-Language Models, Embeddings (Sentence-BERT), FAISS vector search, Python, ANLS, Exact Match, F1 metrics
Team Participant, Trustworthy AI and Cybersecurity Challenges - CISPA, Warsaw, Poland
Duration: 2026
Summary:
- Participated in the European AI and Cybersecurity Hackathon focusing on trustworthy AI and cybersecurity challenges
- The project involved developing methods for dataset inference, LLM memorization extraction, watermark detection, calibration, and robustness analysis under hidden benchmark conditions
Responsibilities:
- Worked on dataset inference and LLM memorization extraction techniques.
- Contributed to watermark detection and calibration methods.
- Performed robustness analysis and failure-mode diagnostics.
- Engaged in architecture-aware statistical inference and contrastive decoding.
Technologies: Large Language Models, Statistical Inference, Contrastive Decoding, Robustness Analysis, Watermark Detection
Instructor - Landmark Certification Centre ML Training Programs
Summary:
- Delivered practical instruction in machine learning concepts including regression, classification, model evaluation, and Python-based workflows
- Guided learners through implementing models and understanding end-to-end training and evaluation processes
Responsibilities:
- Delivered practical machine learning instruction covering key concepts and workflows.
- Guided learners in model implementation and result interpretation.
- Provided end-to-end training and evaluation practice support.
Technologies: Python, Machine Learning (Regression, Classification), Model Evaluation
Enterprise LLM fine-tuning and MLOps
Summary: Implemented parameter-efficient fine-tuning workflows for large language models using LoRA/QLoRA, Hugging Face, MLflow, DVC, and cloud-based experimentation to enable reproducible model development.
Responsibilities:
- Developed fine-tuning workflows for enterprise LLMs.
- Managed reproducible experimentation pipelines using MLOps tools.
Technologies: LoRA, QLoRA, Hugging Face, MLflow, DVC, Cloud Platforms
Conversational Document Intelligence/RAG System
Summary: Developed a document-grounded conversational pipeline using sentence embeddings, vector retrieval, LangChain, and context-aware prompting to answer questions over unstructured content.
Responsibilities:
- Built conversational QA system grounded in document data.
- Implemented retrieval-augmented generation with embeddings and vector search.
- Applied context-aware prompting techniques.
Technologies: Sentence Embeddings, Vector Retrieval, LangChain, Prompt Engineering, RAG
Churn and Retention Analysis
Summary: Built a classical machine learning workflow for churn and retention analysis involving data cleaning, feature engineering, class-imbalance handling, Random Forest modeling, performance evaluation, and feature-importance analysis.
Responsibilities:
- Performed data cleaning and feature engineering.
- Handled class imbalance in datasets.
- Developed Random Forest models and evaluated performance.
- Conducted feature importance analysis.
Technologies: Python, Random Forest, Data Cleaning, Feature Engineering, Class-Imbalance Handling
Computer Vision Projects
Summary:
- Applied computer vision techniques including image classification and object detection using OpenCV, CNN/ResNet architectures, and YOLOv8
- Included transfer learning and model evaluation workflows
Responsibilities:
- Developed image classification and object detection models.
- Applied transfer learning techniques.
- Evaluated model performance and accuracy.
Technologies: OpenCV, CNN, ResNet, YOLOv8, Transfer Learning
Upstaffer since September, 2026
