Jakub Strawa

post-training, RL, continual pre-training, synthetic multimodal data

I share my knowledge writing technical articles on topics like AI, ML, DL, multimodal LMs, LLMs, and AI Agents and many others! https://medium.com/@jakubstrawadev

I focused on advancing research in machine learning and mathematical modeling under the mentorship of Dr. Jarosław Duda. My work included: - Contributing to Hierarchical Correlation Reconstruction (HCR) methodology. - Conducting independent research and publishing a paper on “Improving KAN with CDF Normalization to Quantiles” as the main author. - Currently developing methods to predict statistical moments and reconstruct probability densities using QKAL, exploring novel applications of Kolmogorov–Arnold Networks.

Designed and productionized a medical RAG pipeline (custom retriever, chunking strategy, vector database integration, and evaluation framework). Conducted research into multimodal RAG (vision–language alignment) and knowledge-graph–augmented RAG for structured retrieval and reasoning in complex domains.

As part of a pioneering AI transformation initiative in the banking industry, I contributed to building scalable, secure, and business-critical AI solutions from the ground up. - Led the development of innovative data products using LLMs/GenAI as well as classical ML and Deep Learning techniques. - Researched and prototyped applications of multimodal language models, knowledge graphs, and automated extraction of organizational structures from unstructured data. - Drove sales engineering initiatives by writing and presenting technical PoCs tailored to client-specific challenges.

Productionisation proof of concept ML models of commercial channel optimisation tools on the full range of data, features, and guidelines for more than 20 markets.

Developing a pet-tech ecommerce startup with many new technologies, such as: Digital Twin, Machine Learning