Michał Kordyzon - personal site.
(info about my work)
—————————————————————————————————————————————————————–
Hi there 👋
My new technical blog (it’s a very recent project):mlog
As requested, here is the old link to 2020 Optimization workshops
I am a data scientist working @IBM.
2023 and 2024 were exciting years as GenAI spread rapidly and new breakthroughs appeared almost weekly. During that time, I worked on several GenAI projects using the watsonx.ai (IBM’s platform for LLMs). These projects focused on RAG applications, automated report generation (from both tabular data and documents), and “intelligent” search.
2025 is shaping up to be the year of AI agents. For me it’s importnat. Not only because of their business value, but also as a personal milestone. I’ve been fascinated by agentic AI since 2017, giving public talks and joining early projects, though I still viewed it as a distant future. One example I often used to trigger semi-futuristic, semi-technical discussions was an IBM Research project where an AI agent played Pac-Man. The special thing was that AI agent played ‘ethically’ and avoided ‘killing’ ghosts (the reward function in reinfocement learning). Here is the link: ethical Pacman.
My interests before genAI era :
-
Machine Learning & Data Governance
I engage in my projects as an ML Engineer, or as a Data Scientist. See the notes about my latest ML project as example. In this project I have used tweedie regressor with transformers to count pure premium for insurance product. Model and transformer were hosted on Watson Machine Learning and they used Watson OpenScale monitors to scan bias, explainability, fairness and drift.
-
Natural Language Processing
Currently I am building MVP in area of social listening case, scrapping the content from social platforms, analysing it with python libraries and translating the insights into business value. I am also using Large Language Models from Huggingface, and from IBM watsonx.ai.
-
Optimization
Please see my two Decision Optimization workshops that I made for my clients. Decision Optimization (aka prescriptive analytics) is a great way to enrich machine learning projects. For optimization projects I am using CPLEX, and DOcplex, two IBM libraries that can help build our dynamic programming models efficiently.
-
Visual Recognition
Using Detectron framework on IBM platfrom named Maximo Visual Inspection I have created a model where client’s AGV carts were inspected for security, giving great value of automated daily security checks. Watch this youtube video for more information.
-
Neuro-symbolic AI
It’s always beneficial to attend a good summer school like this one. Neural-symbolic AI is kind of deductive system allowing you to put a reasoning layer on top of your neural nets. You will find general info about neural-symbolic AI here. If you want the hands on experience, take all activities needed to get this badge.
-
Qiskit
In 2020 I have spent a summer (and few consecutive months) learning about Qiskit, brilliant Python SDK for quantum programming. I have contributed to Qiskit in many ways, attended hackathons, and enjoyed being the part of the community. I am glad IBM is making such risky bets. Google’s Larry Page famously said: “Especially in technology, we need revolutionary change, not incremental change.”
Check some of my publications:
- article about Qiskit
- article about Decision Optimization
- System z IBM Redbook that I have co-authored
- System z IBM Redbook that I have co-authored