Christian’s Machine Learning Journey – chrwittm.github.io

Categories

All (50)

German (1)

LLM (2)

ReAct (3)

agent (1)

ai (4)

ai-ethics (1)

anthropic (1)

api (1)

apple silicon (3)

automation (1)

autonomous_vehicles (4)

binding (1)

blogging (5)

btp (2)

c (1)

calculator (2)

chat (3)

chatgpt (1)

claude-code (1)

data (1)

dataset (1)

eda (1)

edge_case (1)

embeddings (1)

fast.ai (8)

fastapi (3)

fsd (1)

function_calling (4)

genai (1)

gpt-oss (1)

grok (1)

groq (1)

hugging face (4)

image_generation (1)

injection (1)

install (2)

javascript (1)

joule (1)

jupyter (7)

kaggle (6)

llama (1)

llama.cpp (2)

llama2 (2)

llm (13)

markdown (1)

math (1)

ml (10)

mnist (2)

nlp (8)

numpy (1)

office (1)

ollama (1)

openai (2)

orchestration (2)

python (3)

pytorch (1)

quarto (4)

rag (1)

render (1)

robotaxi (1)

rpt-1 (3)

sap (7)

satire (2)

tabular (4)

tesla (1)

titanic (3)

tokenization (2)

tools (2)

training (1)

update (1)

vibe-writing (3)

vision (2)

wordpress (1)

x.ai (1)

The State of Autonomous Driving in August 2026

autonomous_vehicles

tesla

fsd

robotaxi

A firsthand report from 1,600 miles with Tesla FSD and two Waymo rides in Northern California.

Christian Wittmann

The Self-Driving Car No AI Could Imagine

autonomous_vehicles

image_generation

edge_case

For my April fools’ article about AI vehicles (original in German / translated to English), I needed a title image that told the whole story at a glance: A self-driving car…

Christian Wittmann

Deutschland seit Jahrzehnten führend bei selbstfahrenden Autos

autonomous_vehicles

satire

vibe-writing

German

Selbstfahrende Autos gelten vielerorts noch als Zukunftstechnologie. Während in den USA und China bereits Robotaxis durch Großstädte rollen, diskutiert Europa scheinbar noch…

Christian Wittmann

Germany has been Leading in Self-Driving Cars for Decades

autonomous_vehicles

satire

vibe-writing

Self-driving cars are still widely considered a technology of the future. While robotaxis are already rolling through major cities in the US and China, Europe seemingly…

Christian Wittmann

Claude Code for Non-Coders: Automating Office Work Through Conversation

claude-code

office

automation

ai

Have you ever wished you could just talk to your computer and have it handle your Office work? Open a PDF, copy the table, paste it into Excel, fix the formatting, copy…

Christian Wittmann

Building Chat with SAP’s Harmonized API

sap

orchestration

llm

chat

function_calling

vision

In Building Chat from Scratch, we created a lightweight chat for Jupyter notebooks. In Turning GPT into a Calculator, we added function calling and the ReAct pattern. And in …

Christian Wittmann

Creating a ‘Hello World’ Joule Agent

sap

joule

agent

btp

ai

Building on Michael Ameling’s excellent Agent Builder in Joule Studio announcement and The Joule Studio Learning Series, this post will guide you through building your first…

Christian Wittmann

Building a Harmonized-API ‘Hello World’ with AI Core

sap

orchestration

llm

The Harmonized API of the Orchestration Service is probably one of SAP’s best kept secrets. It allows you to talk to all the language models on AI core in a harmonized…

Christian Wittmann

Let’s test RPT-1 on Titanic

sap

rpt-1

tabular

titanic

How do you test SAP’s RPT-1 model beyond a simple Hello World example? Its claim to fame is that it can replace traditional machine learning models for classification and…

Christian Wittmann

Building an SAP-RPT-1 ‘Hello World’ with AI Core

sap

rpt-1

tabular

Using SAP-RPT-1 in the playground is a nice way to get a first impression, but you’ll soon realize there are some limitations. In this blog post, let’s explore how to use…

Christian Wittmann

Welcome SAP-RPT-1! What is it? How can you try it out?

sap

rpt-1

tabular

SAP announced SAP-RPT-1 at TechEd 2025 in Berlin. What is it good for, and how can you try it out? In this blog post, I’ll walk you through the announcement and show you how…

Christian Wittmann

Managing Python Environments

python

install

fast.ai

I have a confession to make… Ever since I set up my Mac, I’ve been happily pip-installing all the Python packages I ever needed into the base environment. Foreseeably, this…

Christian Wittmann

Running GPT-OSS Locally on macOS with Ollama + Open WebUI

ollama

gpt-oss

llm

Wouldn’t it be nice to have a ChatGPT-grade assistant running locally on your laptop? That’s now possible with OpenAI’s first open-weight models, GPT-OSS. The 20B version is…

Christian Wittmann

Deploying a FastAPI App to SAP BTP

fastapi

sap

btp

In a previous blog post, we created a simple greeting API using FastAPI, entirely within a Jupyter Notebook. This follow-up post shows how to deploy it on SAP Business…

Christian Wittmann

Deploying a FastAPI App to the Cloud

fastapi

render

In my previous blog post, we created a simple FastAPI-based greeting service with both GET and POST endpoints, an HTML frontend, and interactive documentation. In this…

Christian Wittmann

Implementing a Greeting API with FastAPI

fastapi

jupyter

Creating a RESTful API may sound complicated, but with FastAPI, it’s remarkably straightforward. FastAPI is a modern, high-performance web framework for building APIs with…

Christian Wittmann

The Million-Token Question: What the Bible Teaches Us About LLM Pricing

tokenization

nlp

In my previous blog post, I explored how different languages compare in terms of tokens per word. That analysis was based on Wikipedia articles and raised a few follow-up…

Christian Wittmann

Empirically estimating tokens per word across languages

tokenization

nlp

Tokens are the new currency of generative AI. We’re paying for generative AI usage in tokens, sometimes directly via APIs, more often invisibly when using generative AI…

Christian Wittmann

Beneath the Blue Dome

vibe-writing

Here is a vibe-written short story, inspired by Andrej Karpathy’s idea of “vibe-coding” applied to writing. A few weeks ago, a fly had a short but striking encounter with me…

Christian Wittmann

Implementing Web Search for Large Language Models from Scratch

llm

ReAct

ai

tools

function_calling

One major limitation of large language models (LLMs) is that their knowledge is typically constrained by a fixed cutoff date, beyond which they’re unaware of recent events…

Christian Wittmann

From Intelligence to Generative AI

ai

ml

genai

llm

How is Generative AI related to Intelligence? I’ve frequently seen the following diagram used as an opening slide in AI presentations to set the stage. Presenters, myself…

Christian Wittmann

Why Relative Risk Matters in AI Ethics: A Personal Journey into Gender Inference

ai-ethics

A few days ago, I had a fascinating, and unexpected, journey into the heart of AI ethics. It all started with a conversation about whether it was ethical for AI to infer a…

Christian Wittmann

OmniChat - How to chat with any LLM

chat

llm

python

injection

openai

llama

anthropic

x.ai

grok

groq

So far, I’ve worked with large language models (LLMs) from OpenAI and Meta (the LLaMA series) in separate implementations. I noticed that the code across these…

Christian Wittmann

How to Run JavaScript in a Jupyter Notebook

javascript

jupyter

I decided it was time to learn some JavaScript. Since I’m used to working in Jupyter notebooks for exploratory coding in Python, and my blog is also entirely written in…

Christian Wittmann

Building the Apple Calculator in a Jupyter Notebook

LLM

function_calling

ReAct

calculator

Did you see Apple’s iPad calculator demo on WWDC 2024? Apple calls this feature “Math Notes”. It looks pretty cool how you can hand-write calculations, and it solves them on…

Christian Wittmann

How to Turn GPT into a Calculator

LLM

function_calling

tools

calculator

ReAct

Large Language Models (LLMs) are great writers, but they struggle with numbers: Counting, adding, and basic arithmetic. I guess this is why they are called Large Language…

Christian Wittmann

Updating Quarto and Helpful Features

blogging

quarto

update

After completing the blog post on How LLMs are Trained, I took some time for blog maintenance. A Quarto upgrade was necessary, and I also wanted to introduce you to some…

Christian Wittmann

How LLMs are Trained

llm

nlp

training

Large Language Models (LLMs) like ChatGPT have become invaluable tools for many of us. But what lies beneath the surface of these sophisticated models? How have they been…

Christian Wittmann

Remembering the Wittmann Tours World Trip with RAG

rag

llm

nlp

Back in 2017/2018 my wife and I did a world trip, and we documented it on our blog Wittmann-Tours.de. These 14 month were among the most exciting times of my life, but…

Christian Wittmann

Visualizing Embeddings in 2D

embeddings

llm

nlp

Did you ever try to explain what embeddings are to people who have no or only a limited background in machine learning or computer science? I recently tried this in an…

Christian Wittmann

How to Convert a Wordpress Blog into Markdown

wordpress

markdown

dataset

In this blog post, I will guide you through the steps to convert a Wordpress blog into markdown. While this my seem like a unrelated subject of this blog, it is a…

Christian Wittmann

Building Chat for Jupyter Notebooks from Scratch

chat

jupyter

llama2

llm

Let’s build a light-weight chat for llama2 from scratch which can be reused in your Jupyter notebooks.

Christian Wittmann

How to Implement a C Binding in Python with ctypes

c

binding

llama.cpp

llm

How can compute intensive large language models (LLMs) run on consumer-grade laptops? C bindings are part of this magic, they create wrappers around the C code to make is…

Christian Wittmann

Running LLama2 locally on a Mac

llama2

apple silicon

llama.cpp

hugging face

llm

Running a large language models (LLM), namely llama2, locally on my Mac was the next logical step for me working through the hacker’s guide by Jeremy Howard. While it was…

Christian Wittmann

How to call the OpenAI API from a Jupyter Notebook

openai

jupyter

api

llm

Exploring Large Language Models (LLMs) through their Web-based User Interfaces (WebUIs) is indeed insightful, particularly for experimenting with various prompt engineering…

Christian Wittmann

Running Fast.AI / Huggingface Transformers on Apple Silicon

ml

fast.ai

apple silicon

hugging face

mnist

nlp

In my previous blog post, I described how I setup my Fast.AI development environment on Apple Silicon. In this one, let me share my experience running some notebooks on…

Christian Wittmann

Installing Fast.AI on Apple Silicon

ml

fast.ai

apple silicon

install

I recently moved from Windows to Mac. One of the things which needed attention in making that shift was to re-install everything around Fast.AI.

Christian Wittmann

Titanic with ChatGPT

ml

kaggle

titanic

chatgpt

eda

Reworking Lesson 5, I returned to the Titanic Competition to learn more Fast.AI-concepts. Additionally, I explored how ChatGPT could increase my productivity.

Christian Wittmann

Discovering Disaster Tweet Dataset Limitations

ml

data

hugging face

nlp

kaggle

What started out as a simple exercise to visualize model performance using a confusion matrix revealed that the training set contains lots of incorrectly labeled tweets…

Christian Wittmann

Natural Language Processing with Disaster Tweets

kaggle

nlp

hugging face

ml

Based on the Fast.AI lesson 4, I transferred the approach from Jeremy’s notebook “Getting started with NLP for absolute beginners” to the Kaggle competition “Natural…

Christian Wittmann

Wrapping-up Lesson 3

fast.ai

kaggle

ml

Lesson 3 took me a while to rework, because it churned out quite a few interesting projects:

Christian Wittmann

When disaster strikes: Re-building a Quarto Blog

blogging

quarto

jupyter

For the last 2 months I have been a proud writer of this blog, until yesterday disaster struck: Upon publishing of my MNIST-blog post via the usual quarto publish gh-pages…

Christian Wittmann

MNIST, the ‘Hello World’ of Computer Vision

kaggle

mnist

fast.ai

vision

After Cat vs. Dog, this is the next challenge for me in computer vision: Building on chapter 4 of the book, I challenged myself to implement a model based on the MNIST datase…

Christian Wittmann

My First Kaggle Competition: Titanic

kaggle

titanic

fast.ai

ml

tabular

For more practical experience with gradient descent, I decided to participate in the Titanic Competition. Here is how I did it and what I learned.

Christian Wittmann

Matrix Multiplication

math

python

numpy

pytorch

Since matrix multiplication is a big thing for deep learning and visualizations like http://matrixmultiplication.xyz/ were a bit to fast for me to properly re-understand…

Christian Wittmann

How I created this Blog

blogging

quarto

jupyter

Blogging is an essential part of the Fast.AI methodology, therefore, I decided to follow the advice and start my blog to document my “Machine Learning Journey”. Here are the…

Christian Wittmann

Creating a Blog Post using a Jupyter Notebook

blogging

quarto

jupyter

This is somehow a “Hello World”-notebook, since its only purpose is to demonstrate how you can use a jupyter notebook to write a blog post using Quarto.

Christian Wittmann

Visualizing Gradient Descent in 3D

fast.ai

ml

If you want to understand Machine Learning you have to understand gradient descent, we have all heard that before ;). Since I am a visual person, I tried to not only think…

Christian Wittmann

Fast.AI with Bears, Cats and Dogs

fast.ai

ml

The 2022-version of the Fast.AI course was a welcome and well-timed opportunity for me to continue my machine learning journey.

Christian Wittmann

Hello World!

blogging

Christian Wittmann

Reuse