Letizia Iannucci

pro gradu palkinto

🏆 Honored and humbled by this recognition! Huge thanks to @bolozna.bsky.social and Antonis Matakos for their guidance and support. Excited to keep building on these ideas and push them further 📚

Step 1: 🚨 Wake up to a system failure because API requests are timing out

Step 2: 💻 Drop this gem into slack "I did implement throttling, retries, and sleeping... but probably not enough"

Step 3: 🫠 Realize that this applies to both my code AND my life

Where do you stand on this? Is data science part of IT and engineering, or is it still inside business reporting and analytics? And how do data scientist produce value with AI/ML?

#ai #ml #mlops #leadership #strategy

This is exactly why modern organizations bridge data science with software engineering, and ML engineers fill the technical gap, ensuring AI and ML power applications instead of getting stuck in notebooks or slides. 💻

❓ How much are data scientists expected to know about optimizing models for low-power devices?
🔧 Should they also deliver the entire backend system and cloud infrastructure for an AI application?
📱 What about firmwares, mobile app compatibility, and API scalability?
👀 And what about UX/UI?

💡 Data scientists build AI/ML models that power recommendation systems, forecast revenue, recognize physical activity, and much more. But these models only produce value when integrated into real-world applications, be it your smartwatch, a food delivery app, or an internal business process.

🛵 Your food delivery app has a recommendation system suggesting meals based on past orders, trending venues, etc. This is ML at work. Who delivers it?

⏱️Your smartwatch recognizes running, biking, or sleeping using ML trained on sensor data. How did this become a production-ready feature?

For years, data scientists were placed in analytics teams, producing reports but rarely working with engineers to deploy AI at scale. These silos still exist in many organizations. Is data science really just a reporting function? 🤔

Genuinely curious: where do you see the data science function and why?

📊 Reporting & Analytics Support
or
🛠️ Engineering & Development

#datasky #mlsky #databs

Instead of seeing solving a bug as putting on a band-aid 🩹, I take it as an opportunity to reflect on the system's architecture and goals.

How do you approach debugging and software improvement in your work? 🤔

#softwareDevelopment #softwareDesign #softwareArchitecture #tech #systemsThinking

Beyond "why did this failure happen":

🎯Which part of the codebase allowed it?
📜Why is it coded that way?
👀Is it an unhandled corner case or a consequence of an upstream decision?
❓What should our system do when encountering this or a similar case?

The solution is often in the questions

💭 Every failure is a chance to improve the system.

My favorite technical moments aren't the flashiest 💥.
Sometimes a bug fix is just a few lines of code, but the real work is asking "why" and "what if" (many times).

Pointing out inconsistent or ambiguous abbreviations and var names might feel tyrannical, but it's truly about fostering clarity and quality in the codebase. Conceptual integrity ensures software is reliable, coherent, and maintainable: it's an architectural choice 🚀

#DeveloperVoices #TechWisdom

🌟 "A fundamental architecture practice is defining words, making sure that the definitions of these words are clear. It isn't just about getting along better with people. It'll end up in the code, it'll end up in the database"

Developer Voices is always 🔝

The Data-Conscious Software Engineer The Unicorn That Data Teams Actually Need

The read 📖 that caused this stitch: open.substack.com/pub/dataprod...

Of course CS specializations are still relevant, but a DS should know software design patterns, testing and CI/CD, and a SE should be able to tell supervised from unsupervised learning and to build simple models like linear regression or k-means. Their education does cover these things.

ML/DS/AI are effectively specializations within computer science. People building predictive models go through the same foundational courses as software devs. Why are some organizations not requiring from them the same level of proficiency in the basic concepts?

I've heard of DS siloed in business roles with no access to IT infra, and data teams not having any contacts with SE. Some org might be fine with DS not delivering anything but PoCs and offloading prod deployment to another team. But is "research" and "ad-hoc analysis" a sufficient output? 🧐

For modern data engineers or data scientists, who spend 80+% of their time coding, lacking basic software and devops skills simply means falling short at their job.

At no other time have I learned so much and so fast as while working side-by-side with a "data-conscious" software engineer. SOLID (pun intended 🤓) SEing practices and mindset are must-haves across most roles for a team that aims to ship and maintain production-grade data products.

Thank you for this! I often find myself stressing that "inauthentic != bots". Earlier research also mostly focused on bots, but the landscape of inauthentic behavior is much broader and it’s important to call things by their precise names

Almost every influence operation account is run by actual people who craft their messages individually. They're still inauthentic accounts that are coordinating to manipulate the discussion on a platform in a desired direction or to influence people into thinking a certain way!

Advanced search operators + Google search tricks +Tools to get profile info in one pic

Bluesky search tips

Advanced search operators + Google search tricks +Tools to get profile info in one pic

#osint #socmint

No doubt decluttering is therapeutic, and refactoring is decluttering applied to code (up to a certain scale, or it becomes more like rebuilding a house from scratch) 😊

Have you looked into item-based collaborative filtering? The implementation is pretty straightforward (cosine sim or jaccard sim of items based on users' ratings or buy/no buy choices). However, if the matrix is big, it might be computationally expensive (unless you use some trick)

functional programmers opening Advent of Code day one be like "zipWith!"

🎉 Advent of Code Day One: got my shiny stars ✨✨ by solving the puzzles in Python, Rust, Scala, and TypeScript! 🧩💻

Python 🐍: safe bet
Rust 🦀: rewarding and fast if it compiles
Scala: I super love FP ❤️ but I forgot how messy the Java stacktrace is 🐘
TypeScript: maybe the least interesting for AoC?

Just found out about LocalStack 🤯 It allows to run and test #AWS CDK applications on a local machine (as opposed to my usual synth, push and pray 😂). It could go into CI/CD as well. I feel such a noob for discovering it only now. Anyone has experience with it? docs.localstack.cloud/getting-star...

I use tenacity and nothing to complain about (no databricks in my setup). Besides basics retry strategies, I combine it with loguru to have custom structured logging before or after retry.

For this specific case, a more functional alternative would be `.assign()`, but I still wonder if anyone else is slightly disturbed by this. 🤓 Also, is this all python's fault, since objects are typically passed by reference? Or is it a design choice in pandas? 🐼 Or do I just need a break? 😅