Senior Data & ML Platform Engineer | dutchengineer.substack.com | Host of Ask A Data Mentor Podcast

Built real-time ML systems delivering millions of daily inferences. Integrated LLMs that saved over 1,000 hours per week across the company. Designed scalable MLOps platforms that took model deployment from months to days.

I have spent the last 8+ years driving AI and ML engineering efforts across startups and high-growth companies. My work spans productionizing models, leading company-wide ML initiatives, and building systems that work. 

Highlights:
Led LLM integration at Equipmentshare, embedding OpenAI into internal tooling to save 1,080+ hours weekly
Deployed Zwift’s real-time racing score model, handling 8.6 million inferences per day
Cut ML deployment times from months to 3 days with standardized CI/CD and packaging
Rolled out Databricks company-wide, reducing deployment cycles by 4x
Built ML pipelines for pricing $3B+ in fleet assets, churn risk, and invoice automation

I write often on machine learning systems and infrastructure on dutchengineer.substack.com, and I enjoy helping teams ship fast, reliable ML in production.

Keywords: Airflow, AWS (Cloudwatch,  ECS, EKS, Fargate, Glue, Redshift, Kinesis, Databricks, Postgresql, S3), Azure (Azure Functions, Azure ML Studio, Databricks), BentoML, C++, Datadog, Docker, ElasticSearch, Flink, Kafka, Keras, Kubernetes, Hive, High Performance Computing, Github Actions, Gitlab CICD,  Pagerduty, Prefect, Python (DBT, Numpy, Pydantic, Sklearn,Scipy), Pyspark, Neural Networks, Natural Language Processing, MLFlow, MongoDB, Recommendations, Sagemaker, Scala, SparkML, SQL, Tableau, Terraform, Pytorch, GPU

Sarah Floris, MS

Senior Manager - Data Science and Engineering at Apple

I lead the data science and engineering team at Apple Finance - Services and Infra. The team focuses on translating business problems and requirements into data products using advanced statistical and machine learning models to support and help Apple leadership make data-driven decisions.

I am also an open-source contributor and the author of Hands-On Time Series Analysis with R (Packt publication) and several R packages for time series analysis and machine learning applications. Currently working on my next book, Applied Time Series Analysis and Forecasting, which focuses on forecasting at scale. 

Also available on:
Github: https://github.com/RamiKrispin
Medium: https://medium.com/@rami.krispin
Mastodon: https://mstdn.social/@ramikrispin

Rami Krispin

Imagine going to a restaurant and starting with dessert—then ordering the main course.

That’s roughly how university degrees are structured today, in my opinion: a heavy focus on theory, with far less emphasis on the practical tools students actually need in the job market.

Personally, I believe the very first course in any STEM track should be tooling—Git, the command line, shell, Bash, and related fundamentals. Having this foundation early would allow students to gain real, hands-on experience throughout their degree, not just after graduation.

A few years ago, three MIT PhD students recognized this gap and created a course called “The Missing Semester.” The course focuses on essential practical skills like the CLI, Git, Vim, and shell tools. It first launched in 2019, ran for two years, and last week the original instructors returned to MIT to teach an updated version.

The full course is available on YouTube:
https://lnkd.in/gh4KRiD8

Highly recommended—especially if you’re at the beginning of your journey.

#datascience #ai #mlops

Senior Manager - Data Science and Engineering at Apple | Docker Captain | LinkedIn Learning Instructor

Hi Bay Area data folks! 👋

Bay Area Data & AI is hosting an in-person, casual networking event focused on data, AI, careers, and products.

I’m excited to co-host this event with Manisha Arora, Devshree Bharatia, Vishakha Sadhwani,  Venkata Naga Sai Kumar Bysani, and  Sriram Arumelli.

📅 When: Jan 24 (Sat) at 10:00 AM PST
 📍 Where: Sunnyvale Baylands Park

👉 More details & RSVP: https://luma.com/kgsx6o61

Hope to see you there!

#ai #data #datascience

How to make your Python Docker container slimmer 🐳👇🏼
 
If you’ve ever deployed Python containers only to watch them balloon to several GBs, you’re not alone — I once hit a 5 GB Python image (thank you, PyTorch dependencies 😅). But with a few simple optimizations, you can dramatically reduce image sizes, speed up deployments, and lower cloud costs.

Here are practical tips that have worked well:
1️⃣ Choose a Lean Base Image
Start smaller:
 ✔️ python:3.x-slim instead of the full Python image
 ✔️ python:3.x-alpine if compatible with your dependencies (and you are a brave developer 😆) 
 Smaller base = smaller final image and faster pulls. 
2️⃣ Use Multi-Stage Builds (Big Win!)
Multi-stage builds let you install tools and build dependencies in one stage, then only copy what’s needed into a clean final image. This can reduce the size by 60-90% compared to single-stage builds. 
3️⃣ Minimize Layers
Every RUN / COPY / ADD adds a layer.
 ➡️ Combine related commands with && to reduce layers and eliminate intermediate files (although you won't gain much size reduction).
 Example:

RUN apt-get update && \
    apt-get install -y build-deps && \
    pip install -r requirements.txt && \
    apt-get purge -y build-deps && \
    rm -rf /var/lib/apt/lists/*

Fewer layers = smaller image. 

4️⃣ Use .dockerignore to Cut Build Context
Exclude unnecessary files (.git, logs, IDE folders, virtual environments) so Docker doesn’t send them into the build context — this keeps images lean and speeds up builds. 
5️⃣ Install Only What You Need
Separate dev dependencies from production deps.
 Use flags like --no-dev or careful requirements files so you install only what your app actually needs at runtime. 

💡 Why it matters: smaller images → faster CI/CD → less storage & network cost → reduced attack surface → happier developers.

If you want to learn more, please check my tutorial about multi-stage build:

https://lnkd.in/gHTpdt2S

#mlops #docker #devops #python

Open article: Introduction to Multi-Stage Image Build for Python by medium.com

Introduction to Multi-Stage Image Build for Python

Introduction to JAX and Flax NNX - the framework behind Google Gemini 🚀👇🏼

JAX is a Python library for accelerator-oriented array computation and program transformation, designed for high-performance numerical computing and large-scale machine learning. Flax NNX is a neural network library and ecosystem built on top of JAX.

The following tutorial from Google for Developers provides an introduction to the JAX AI stack, and it covers topics such as:
✅ JAX’s functional transformations
✅ Debugging techniques
✅ Scaling models across distributed hardware
✅ Optax for optimization and checkpoints

📽️: https://lnkd.in/g8kfrFmg

#ai #llm #machinelearning

My weekly newsletter is out! 

This week’s agenda:
🔹 Open Source of the Week - Pandas 3 release
🔹 New learning resources - Introduction to JAX and Flax NNX, OpenCode crash course, introduction to FastAPI
🔹 Book of the week - Introduction to Machine Learning Systems by Prof. Vijay Janapa Reddi

📌 Subscribe to receive weekly updates: https://lnkd.in/gb3P8YdE
🎯 Learn how to build an SQL AI agent: https://lnkd.in/g4_W8na8

#python #datascience #machinelearning #ai

This week's agenda: Open Source of the Week - Pandas 3 release New learning resources - Introduction to JAX and Flax NNX, OpenCode crash course, intro

Open article: Pandas 3 is Out, Introduction to Machine Learning Systems, New Tutorials | Issue 72 by Rami Krispin • 3 min read

Pandas 3 is Out, Introduction to Machine Learning Systems, New Tutorials | Issue 72

My weekly newsletter is out! 

This week’s agenda:
🔹 Open Source of the Week - The SQLGlot project
🔹 New learning resources - Setting Memory for AI application, MIT deep learning course, Statistical Rethinking 2026, introduction to AI engineering, LlamaIndex crash course
🔹 Book of the week - Statistical Analysis with Python For Dummies by Joseph Schmuller

📌 Subscribe to receive weekly updates: https://lnkd.in/gb3P8YdE
🎯 Learn how to build an SQL AI agent: https://lnkd.in/g4_W8na8

#ai #python #sql

This week's agenda: Open Source of the Week - The SQLGlot project New learning resources - Setting Memory for AI application, MIT deep learning course

Open article: Statistical Analysis with Python, SQLGlot & MIT Deep Learning | Issue 71 by Rami Krispin • 5 min read

Statistical Analysis with Python, SQLGlot & MIT Deep Learning | Issue 71

Many of the AI concepts are simpler than they seem. Adding to an AI application's "memory" is one of those things. 

By default, LLMs are stateless, meaning each conversion is independent from the previous ones. Here is how you turn your AI application with less than five lines of code:
1️⃣ Set an empty list to store the previous conversions
2️⃣ Add to the prompt template this list as an argument
3️⃣ After every response of the LLM update the list with the user question and LLM response

This is, of course, not a production-ready method. There are a few limitations to this method, such as an increase in the number of tokens used and a potential drop in the context as the conversion takes a different direction.

The slides below, generated with NotebookLM, summarise this process and are based on a tutorial I created this week. The next tutorial will focus on efficient management of the token size.

This tutorial and others are available in the AIOps Newsletter: https://lnkd.in/gghdpeEF

#ai #llm #datascience

Building Memory to LLM Applications

I share an 𝐨𝐩𝐞𝐧-𝐬𝐨𝐮𝐫𝐜𝐞 project every week in my 𝐧𝐞𝐰𝐬𝐥𝐞𝐭𝐭𝐞𝐫. This week’s focus is on the SQLGlot project 👇🏼

SQLGlot is a versatile library for reading, manipulating, formatting, and translating SQL across many dialects. It ingests SQL text, generates an abstract syntax tree (AST), and produces syntactically correct SQL for a target dialect such as DuckDB, Spark, Snowflake, BigQuery, Postgres, MySQL, and many others. It also includes features for query analysis, syntax error detection, and basic execution.

Why It’s Useful 🎯
✅ Dialect Translation: Simplifies migrating SQL across engines by automatically transpiling queries.
✅ Parser & Formatter: Can parse diverse SQL inputs, ensure correct syntax, and produce clean, standardized output.
✅ AST-Level Manipulation: Lets developers programmatically inspect and modify query structures, enabling tooling such as linters, analyzers, and automated refactors.
✅ Optimizer & Analyzer: Includes basic optimization rules and expression tree traversal utilities.
✅ Execution Engine: Although not designed for performance, SQLGlot can execute queries against Python data structures, useful for testing or in-memory analytics.

Key Features 🛠️ 
🔹Transpile between 30+ SQL dialects (including Postgres, Spark, Hive, Snowflake, BigQuery, SQLite, TSQL, and more).
🔹Detailed parser errors & warnings with context.
🔹Extensible dialect system, so you can plug in support for custom SQL syntaxes.
🔹 AST diffing, helping compare and transform queries semantically.
Benchmarks show performance that’s competitive with other Python SQL parsers.

More details are available on the project repo: https://lnkd.in/gb9UxetB

📌 Join 34k subscribers to receive weekly updates: https://lnkd.in/gb3P8YdE

VScode Dev Containers Extension 🐳👇🏼 

One of the main reasons I started to use VScode was the Dev Containers extension. This extension enables the seamless launch and development of a project inside a container.

Here are the main advantages of this extension:
✅ It is highly customizable and flexible provide you with complete control over your project settings. 
✅ Issolation - From the environment setting to the fonts and theme, it enables you to isolate your project VScode settings from your regular settings
✅ Reproducability - beyond your container, it enables you to reproduce your complete settings from fonts to extensions. This is great when you are working on multiple machines and cross-collaborating with others 
✅ Consistent environments — All developers use the exact same OS, tools, runtimes, and libraries. No more “works on my machine” problems. 
 ✅ Faster onboarding — New team members can be productive in minutes without manual setup. 
 ✅ Infrastructure as code — Dev container configs live with your project and can be version-controlled. 

Pro tip 🎯
Set GitHub templates for different use cases, for example, I have the following templates:
Python: https://lnkd.in/gSTcBbcK
R: https://lnkd.in/gKzaBC7T

If you want to get started with the Dev Containers extension, please check the following tutorial:
https://lnkd.in/gtCtrBqM 

#docker #mlops #datascience

Open article: Getting Started with the Dev Containers Extension by medium.com

Getting Started with the Dev Containers Extension

taplio

The Battle

Sarah Floris, MS

Rami Krispin

Followers

Posts

Engagements

Likes

Performance ScoreRami Krispin's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Sarah Floris, MS

Rami Krispin

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Rami Krispin's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreRami Krispin's LinkedIn Performance