Lorenzo Pasqualis

A recent K8s podcast with Clayton Coleman and Rob Shaw is a fascinating listen: https://lnkd.in/gJmpX_DB Why do standard K8s approaches don't work with LLM inference? Unlike a normal web app, LLM generates text one token at a time, each one depending on the one that came before. The technique to remember the context of the conversation is called KV caching. Now regular K8s tools for balancing traffic, adding servers and managing resources at huge scale work really well, because the amount of required varies depending on the prompt. The tools the community created to solve this problem: 🛠️  vLLM - popular inference server that is very good at KV cache management. 🛠️  Inference Gateway - smarter load balancer that enables efficient traffic distribution. 🧩 The llm-d project brings these together in a ready-to-use stack, tested and benchmarked well-lit paths, smarter routing, intelligent work splitting to optimize time-to-first token, methods to run "Mixture-of-Experts" models across multiple servers

50

Learning science principles underpin all the educational tools we create. The stronger those are, the better our tools can be. That’s why I’m so excited to share that we’re infusing #LearnLM into Gemini 2.5, making it the world’s leading model for learning.  This update expands the potential of what learners can achieve with richer, more complex and useful capabilities to help anyone learn. Here are a few of the updates we’re thrilled to share today: NotebookLM: You can now control what kind of Audio Overview you want to hear, such as a quick, high-level overview, or an in-depth explanation. Search: AI Mode lets you ask your toughest questions right in Search, and get high-quality information you can easily explore. Gemini app: Turn your study materials and notes into customized quizzes that provide helpful hints, summaries, and even explanations on why an answer is right or wrong. Learn About: Now delivers explanations with greater, more relevant detail—even letting you upload your own documents so outputs can be framed around your materials, notes, or research papers. We’ve also expanded access for teens and added session history so you can keep track of progress. This is just the beginning of a new chapter of learning, where experiences become more personalized, engaging, and impactful than ever before.   More details in the blog post here! https://lnkd.in/ezJhQJHq

82

1 Comment

In today’s Substack post from the Plan of Record series, I walk through the operational frameworks I’ve relied on to build high-performance product development teams and why they’re essential for diagnosing and fixing what’s broken. I also explain why we had to create the PoR Prioritization Framework to address a critical gap no other model could fill. Here’s the link: https://lnkd.in/ek9J_cqq #SaaS #ProductManagement #EngineeringLeadership #R&D #Prioritization #Execution #CPO #CTO #CPTO

14

“Why are we using a rocket launcher to butter toast?” you might ask... What happens when a perfectly sensible architectural decision quietly drifts out of alignment with reality. A SaaS search engine built for full-text search and clever relevance scoring… being used for glorified key-value lookups. Seventy-five million requests a month. Lots of cost. No edge caching. Limited observability. Ouch. The real lesson here isn’t just “this technology is faster and cheaper” (although 35ms P50 latency and a 0.0001% error rate are nothing to sniff at). It’s about architectural honesty. Stripping a system back to what it needs. Using organisational learning instead of reinventing the wheel. Designing observability in from day one. And migrating gradually, with dual-running and rollback paths, so 75 million monthly requests don’t even blink. 📣 Sometimes the strategic move isn’t adding something new. It’s removing the unnecessary. We’re back to our technical roots this week with a tale from Nick Theodoulou about re-architecting for simplicity https://lnkd.in/eGh_4fU5

17

9 Comments

We just published a short write-up on how teams are using Edge Delta’s Service Map to zero in on hard-to-trace issues—illustrated with a real payment-flow incident that was resolved in minutes once the map highlighted the failing service. If you’re curious about practical ways to cut through log and metric noise, the post is worth a look. Credit to Emrah Samdan for steadily driving this capability and the customer feedback loop behind it. Read here → https://lnkd.in/g3krzkaN #observability #kubernetes #devops #servicemap

39

1 Comment

In recent years, the United States has seen a dramatic increase in book challenges and bans, and Utah has become one of the most significant case studies in this national trend. In 2024, Utah enacted HB 29, allowing the removal of books statewide if just three school districts (or two districts and five charter schools) label them as containing “objective sensitive material.” By 2025, 18 books, including works by Judy Blume, Margaret Atwood, and Rupi Kaur, had been banned from every public school in the state. While the stated intent is to protect children, the real impact is much broader. When we remove books that explore complex topics—identity, empathy, injustice, or resilience, we also remove opportunities for students to think critically, connect emotionally, and see themselves in stories. According to the American Library Association, 2024 saw over 6,800 book bans nationwide, the highest number on record. Yet data consistently shows that access to diverse reading materials directly supports higher literacy rates and stronger emotional intelligence. Banning books doesn’t just silence authors, it silences readers. It narrows the world our children get to read about, imagine, and learn from. As educators, parents, and readers, we can model something better: trust in curiosity, open dialogue, and the power of story.

12

I thought this was a good and very reasonable talk by Eleanor Millman on how to prioritize projects for a platform engineering team: https://lnkd.in/gaYtUHRH A few insights I took from it: - Unlike product features which can be more directly tied to revenue, platform engineering work is a couple steps removed. But that doesn't mean it's unimportant! Far from it, platform engineering projects can make everyone at the company more efficient, able to produce higher quality software, etc. - While "urgency" is always going to be a factor, you don't want to just be putting out fires, you want be able to prioritize work which is "high impact" ... even if that impact isn't felt immediately. - Good rule of thumb: prioritize "highest impact for lowest effort". - The talk has some ideas on what factors to choose for impact, and how to blend them. For instance "speed of development" is one factor, "cloud cost optimization" might be another. The weighting of different impact factors can change over time, depending on the needs of the business. I would say that a lot of companies don't have much methodology here but I can really see the value in codifying it. You can always change the methodology or the weighting if it's spitting out results that don't pass the smell test or you really feel it is leading the company astray. Clarity can give the org more freedom to put resources behind projects that would otherwise never be taken on. When things are unclear, prioritization still happens implicitly, but the decisions tend to be a lot more random and fear-based and the org is worse off as a result.

15

3 Comments

Some random thoughts as we start what is sure to be an exciting and chaotic 2026: - The word "slop", describing AI generated content, is a great word of the year for 2025. - A bullseye quote from a friend of mine the other day: "The only person who likes ChatGPT generated content is the person who generated it." - AI assisted code (Claude Code, etc) has shown how easy it is to write code, but also that actually writing the code isn't the hard part of being a Software Engineer. The hardest and most important part about being a Software Engineer is the thinking (ideation, problem solving, delivering value). - Vibe Coding is a great phrase, and my choice for "word" of the year for 2025. If you don't like it, remember it is not "Vibe Thinking" rather it describes the creative ideation process where the code is generated while in a "thinking flow". - Every Instagram video post comment thread is littered with "AI" comments. Sigh... And a few random non-AI thoughts that have been bouncing around in my head the past couple weeks: - Why was Snoop Dog suddenly everywhere in 2025? Is he still relevant? Why? I don't get it, obviously. - It's been 45 years since the NBA introduced the 3 point line and to me the NBA is nearly unwatchable, unless you like watching 3 point chuckers and no rebounding. Bring back the 60's, 70's or 80's style! Perhaps only enable the 3 point line in the final 2 minutes of each half. - Prediction: Drake Maye and the New England Patriots will win the Super Bowl this year. Hope everyone has a great 2026! No AI was used to create this slop, er, post.

35

8 Comments

I’ve been thinking a lot about accountability—and how misunderstood it often is, especially in fast moving teams. Too often, it gets lumped in with blame and negative consequences after things go wrong. But that’s not how I see it. Real accountability is about clarity and ownership. To hold someone accountable means to be sure there is clarity in expectations and objectives. It’s knowing what success looks like, what’s expected, and why it matters. When managers and engineers have that kind of clarity—and the support to get there—they rise to the challenge. Not because someone’s watching, but because they believe and care. Without clear goals or a shared definition of “done,” and without the right amount of empowerment, accountability isn’t just ineffective—it’s unfair. The best teams I’ve worked with aren’t afraid of being accountable. They want to own things, to ship great work, to know where they stand. But they can only do that when we set them up for success: with the right context, tools, trust, and space to build. As leaders who like to make accountability part of our culture, we must provide all these. Accountability done right isn’t pressure—it’s pride. Would love to hear what you think.

51

8 Comments

Aside from going deeper into the Solana (SVM) stack engineering systems and breaking its performance bottlenecks, a portion of my work right now at Sonic is centered around: Tightening the Feedback-Iteration Loop. Here's how it's going and why I decided to focus on this: Through the years, I've realized that building a startup is more about intelligently probing opportunities and gathering as much data as possible. It's not about perfection. It's not about flashy tech. It's not about hype. It's about getting as close to the truth about our target customers as possible. And what's the best way to achieve this? Keep launching shit and see how people react. Until one sticks where we can triple down and go deeper. This is why tightening the Feedback-Iteration Loop matters. If I can systemize and tighten the Trial and Error loop, we can swing much more than if we let the loop stay loose. #blockchain #engineering #solana #rollup #distributedsystems #protocol

15

Interesting thoughts about AI, episode 1. What will AI do to standard Agile practices and efficiency metrics? For years we've been honing the idea of a story, talking about the smallest possible business value, making tracer bullets, steel threads, etc. We've been optimizing our processes for minimal story cycle time. (Well, some of us have been!) If AI coding agents are so much faster and better, and can make much larger changes at once, this turns some of the Agile practices upside down. Planning / Grooming turns into prompt development Stories might get larger, or not necessary Pointing is redefined, and not slowly Metrics go off the charts.. or are useless what else? I've had a couple of these types of thoughts or realizations lately. Join me next week for episode 2.

13

1 Comment

Learning and Leading with AI at Vista BPSS 🚀 Day one and two at the Vista BPSS has been nothing short of inspiring. Here are a few standout takeaways: 🔹 AI agents are no longer future talk—they’re here, and they’re solving mission-critical problems. 🤖 From automating decisions to transforming workflows, they’re reshaping how we deliver value. 🔹 Design AI agents with results in mind. It’s not just about automation—it’s about building systems that drive measurable business outcomes. 🔹 Adopt an AI-first mindset—while keeping the lights on. ➡️The challenge (and opportunity) is leading innovation while sustaining performance. The balance is key. Excited for what’s next and how these insights will shape how we lead in tech and product delivery. #AILeadership #BPSS2025 #EngineeringExcellence #VistaEquity

13

In typical deployments, a single DAI-50 enables manufacturers to reallocate approximately 1–2 FTEs from manual inspection to higher-value work. Subsequently, this results in: 1./ Less hiring and retraining churn 2./ More time spent on high-value QA projects 3./ Higher employee satisfaction and retention For those interested, here's a case study that goes into the full array of benefits AVIS provides manufacturers. https://lnkd.in/dRj-SmHW

19

1 Comment