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

I create accessible data science content that reaches over 100,000 monthly readers through my 180+ Towards Data Science articles and 800+ daily CodeCut tips covering Python, data workflows, and ML engineering. 

My mission is to make open-source tools approachable through clear tutorials, documentation, and videos while helping data scientists adopt engineering best practices from version control to deployment.

Khuyen Tran

What if you could shrink your Docker images by 30x without touching your Dockerfile?

Docker images include the entire OS layer. For a simple Python app, this is unnecessary because it never touches the shells, compilers, and system utilities bundled inside.

This inflates images to hundreds of megabytes, wasting storage and adding time to every deploy.

Slim automatically analyzes your container at runtime to identify which files are actually used, then builds a minimal image with only essential components.

Slim works alongside Docker, not instead of it:
• Step 1: Build your image with "docker build"
• Step 2: Minify with "slim build your-image"
• Step 3: Push the ".slim" image to your registry

Your Dockerfile and workflow stay the same.

Plus, Slim is open source! Install it with "brew install docker-slim". 

📬 Want more production-ready data science tips? Subscribe to my newsletter at https://bit.ly/49sglaM

#Docker #DevOps #Containers #Python

Author of Production-Ready Data Science | Founder @ CodeCut

Do you trust a package released two hours ago in production?

Installing updated package versions is essential to benefit from new features and bug fixes.

However, freshly released versions can introduce bugs or incompatibilities before the community has time to catch them.

uv from Astral's exclude-newer option lets you set a cooldown period to skip packages released within a specified timeframe.

To use it, add exclude-newer = "7 days" to pyproject.toml and customize the duration as needed. 

📬 Want more production-ready data science tips? Subscribe to my newsletter at https://bit.ly/49sglaM

#uv #Python #DevOps #PackageManagement

What if you could extract web data just by describing what you want?

Traditional scraping with BeautifulSoup follows a familiar pattern: fetch HTML, inspect elements in DevTools, and write CSS selectors to extract your data.

But websites don't stay static. When the HTML structure changes, your selectors break, and you're back to rewriting code.

ScrapeGraphAI uses LLMs to extract data from natural language descriptions. Simply describe what you want in plain English, and the LLM figures out the extraction logic automatically.

Key features:
• Self-healing scrapers that adapt when websites are redesigned
• Type-safe output with Pydantic schema validation
• Built-in JavaScript rendering for React, Vue, and Angular sites
• Multi-page scraping with SearchGraph for research tasks
• Cloud or local models via OpenAI, Anthropic, or Ollama

Plus, ScrapeGraphAI is open source! Install it with "pip install scrapegraphai". 

---
🚀 Full article on ScrapeGraphAI: https://bit.ly/4sZh0YW 
☕️ Run this code: https://bit.ly/4jU4TIo 

#Python #WebScraping #LLM #DataEngineering

What if you could experiment with your AI prompts as easily as changing a config file?

Finding the right prompt often takes experimentation: tweaking wording, adjusting tone, testing different instructions.

But with prompts hardcoded in your codebase, each test requires a code change and redeployment.

MLflow Prompt Registry solves this with aliases. Your code references an alias like "production" instead of a version number, so you can swap versions without changing it.

Here's how it works:
• Every prompt edit creates a new immutable version with a commit message
• Register prompts once, then assign aliases to specific versions
• Deploy to different environments by creating aliases like "staging" and "production"
• Track full version history with metadata and tags for each prompt 

📬 Want more data science and AI tips? Subscribe to my newsletter at https://bit.ly/4p4WfZX

#MLflow #LLM #GenAI #PromptEngineering

What if you could extract any information from text with just one example?

Named entity recognition extracts entities like names, dates, and organizations from text.

But pre-trained NER models can fail on domain-specific text. They weren't trained on medical terms, so "Metformin 500mg" gets labeled as "LAW" instead of "medication".

Fixing this means retraining with thousands of labeled examples.

LangExtract is Google's LLM-powered extraction library that skips retraining entirely. It works on any domain with just one example.

Plus, every extraction includes:
• Exact character positions for source verification
• Attribute grouping to link related entities
• Interactive visualizations to review results 

☕️ Run this code: https://bit.ly/3LY4TKZ 
---
📬 Want more data science and AI tips? Subscribe to my newsletter at https://bit.ly/4jRpfSI

#Python #NLP #LLM #MachineLearning

Find min/max rows by another column in one line with Polars v1.37.0 🍰

Finding the row with the minimum or maximum value based on another column requires sorting, grouping, or complex filter expressions.

Polars v1.37.0 adds min_by and max_by expression methods. These methods find minimum or maximum values based on any column in a single, readable expression.

Get min_by and max_by by upgrading: "pip install -U polars". 

☕️ Run this code: https://bit.ly/49EPy

—
📬 This post is part of the CodeCut newsletter, which focuses on data science, AI, and open-source tools. Subscribe here → https://bit.ly/4p4WfZX

#Python #DataEngineering #DataScience

taplio

The Battle

Sarah Floris, MS

Khuyen Tran

Followers

Posts

Engagements

Likes

Performance ScoreKhuyen Tran's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Sarah Floris, MS

Khuyen Tran

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);}Khuyen Tran's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreKhuyen Tran's LinkedIn Performance