Hello 👋, I am Andrejs

Welcome to my digital garden 🌱 - a mix of atomic notes connected topographically (like Wiki), rather than chronologically (like blogs). This place is an assumed “second brain” and includes study notes, ideas and anything else I found useful to write down.

The domain fluentnumbers was born from the fact that I am fluent with numbers, but also from my strong conviction that numbers (as in “data”) can be fluent and say a lot about a business or engineering topic at hand. And I sure enjoy doing the translation!

This space is constantly in development (except when it is not), hence imperfect in nature. Links without destination mean I would want to write something there, but so far haven’t. There is Explorer ⬅️, Search bar ⬆️ as well as Graph view ⬇️ to help you find stuff.

About me

I am a data scientist and ML engineer on a continuous journey of learning by doing & sharing. I live in the Netherlands and have 10 years of experience in Machine Learning for various industrial and scientific applications including remote monitoring, biomedical algorithms, predictive maintenance, digital twin, asset management, etc.
Besides that I

• try to read more, because a book constitutes a means of transportation through the space of experience, at the speed of a turning page
• enjoy somewhat longer walks and hikes as well as travelling places to see how people live elsewhere
• lately, devote a decent amount of time to Mafia (or Werewolf) games, but then in its competitive (not party) form

---

CONTACTS AND LINKS

• Email
• LinkedIn
• Medium
• ORCID

PROFESSIONAL EXPERIENCE

2024 — …
Quant Researcher at Shell Asset Management Company, The Hague, the Netherlands

2022 — 2024
Data Scientist at Philips, Eindhoven, the Netherlands

2016 — 2022
Senior R&D Engineer at Imec, Eindhoven, the Netherlands

---

Selected notes

Latest

• Enterprise RAG patterns
• what can go wrong with LLMs
• how to evaluate LLM chatbots
• Briefly about transformer’s evolution or why is softmax cool

Selected LLM notes

selected LLM notes

Hot topics in LLM

LLM - basic info and the source to other notes for deeper understanding
LLM inference basics and inference optimization
Paper Review - Llama 3.1 model - how is the one of the best open-source models designed?

Tokenizers, embedding models - corner stone of each LLM task and Byte Pair Encoding as one of popular tokenization ideas
Scaling Laws - how to distribute compute budget efficiently between dataset size and number of trainable parameters?

RLHF - how reinforcement learning finally found its application
Reward Modeling - how it is used for RLHF and LLM fine-tuning?
Direct Preference Optimization - is it a more stable and cheap alternative to RLHF?

What is RAG? How does it work and how to maximize its impact?

Deep Learning architecture blocks

Transformer architecture and how each of the following blocks fit together
Attention, self-attention. multi-headed attention
Skip connection
Batch normalization
Layer normalization
Dropout
Gated Linear Unit
Briefly about transformer’s evolution or why is softmax cool

Evaluation of LLMs

LIST
FROM outgoing([[model evaluation]]) OR [[model evaluation]] AND -[[]]
WHERE contains(parent,[[model evaluation]])
SORT file.name ASC

---

Link to original