David Ward

The INFORMS Analytics Framework is now officially available for download! https://lnkd.in/efQDFN4d This comprehensive framework offers a robust checklist, guiding practitioners through every crucial stage of an analytics project—from the initial framing of the business problem to data collection, model development, deployment, and ongoing maintenance. It embodies industry best practices designed to ensure the successful execution of applied analytics initiatives. It was a true honor and pleasure to contribute to the development of this essential resource. This framework also serves as the foundational bedrock for the Certified Analytics Professional (CAP) exams, now offered at Essentials, Professional (coming soon), and Expert levels. Crucially, the principles embedded within this framework are directly applicable to AI projects. I believe this significant contribution from the INFORMS community, built upon decades of experience in delivering successful projects, deserves wider awareness and adoption within the AI community.

79

4 Comments

https://lnkd.in/gzNqdKHW 2024 was a stellar year for the All of Us Research Program from a scientific perspective, but from a budget perspective it was challenging. The legislation that funded our program for 10 years is tailing off, and we're more dependent on annual congressional funding. That funding did not come through last year and we had to make substantial cuts to the program. We're still waiting on our budget for 2025, so we've had to take deeper cuts. We are now operating with a budget that is 29% of our 2023 budget. This has stressed us on all fronts, but we're hopeful that congress will restore our budget so we can continue to engage underrepresented communities, enroll children, conduct longitudinal data collection, and release new data for research and discovery. The program has been a fantastic investment so far. For example, researchers using All of Us data have discovered more than 275 million previously unreported genetic variants, providing researchers new pathways to better understand the genetic influences on health and disease. All of Us researchers have improved precision medicine for certain cancer treatments, optimized genetic tests to more accurately predict disease risk for 10 health conditions, and identified different health problems linked to subtypes of type 2 diabetes.

66

2 Comments

Everyone wants to be more "proactive" and less "reactive," but it's hard to predict how traffic patterns will evolve or how increased utilization in one service might impact other services. Check out this article from Endre Sara for what to consider when trying to build more resilient applications.

29

1 Comment

"Why Some Organizations Still Embrace Legacy Systems... A thought-provoking observation from the field: Many enterprises aren't holding onto legacy systems just out of habit or fear of change. They're making calculated decisions based on legitimate concerns about modern alternatives: Performance isn't always better in newer systems. Sometimes simpler, purpose-built legacy solutions outperform their feature-rich successors. Modern systems often come with built-in tracking and telemetry that organizations may not want or need, raising privacy and data governance questions. Mandatory automatic updates can disrupt operations and remove control from IT teams who need predictable system behavior. Update mechanisms, while convenient, can create security vulnerabilities and potential backdoors that expand the attack surface. The shift from ownable software to vendor-dependent services means less autonomy and potential risks if vendors change direction or pricing. These aren't just theoretical concerns - they're real factors shaping enterprise architecture decisions. What's your take on balancing modern capabilities with organizational control? #EnterpriseIT #SystemArchitecture #TechnologyStrategy

127

47 Comments

Missed checks? Unlogged issues? Half-done tasks? It’s not a lack of effort... it’s a lack of visibility. When handovers rely on memory or sticky notes, things slip. EviView makes shift transitions: 1.  Clear 2.  Trackable 3.  Consistent No more guesswork. Just smooth, reliable handovers.

20

What happens when you square root technology -- you get it's base root: tékhnē -- it also has a mythic side, brewed from a story with Prometheus 🔥 I learned about the meaning and it's connection from David Dominguez Hooper, from our recent episode. David is building ELDÆON (https://www.eldaeon.com/), the world’s first tactical UAP-detection network -- probably the first uap-first startup that I know that is raising 🚀 I will leave with a quote from David: “Technology is your calling—augmented by tools given by the gods.” Long form of podcast: https://lnkd.in/gnfaRYEQ

21

Ground Truth Development. I’ve been thinking about that term for a while. There’s been a lot of attention on evals in GenAI systems, how we measure quality, alignment, and effectiveness. At the center of it all is human-labeled data. Not just any data, but ground truth, the final authority on whether the model is right or wrong. In AI, ground truth is the canonical reference for evaluation. If it’s inconsistent or poorly managed, everything downstream becomes unstable, like software built on unreliable state. And yet, there’s often a casual assumption that ground truth “just exists”. But where does it come from? Who creates it? How is it maintained and reviewed? These aren’t trivial questions. The expectations for ground truth are far higher than for training data. Ground truth doesn’t just inform the model, it judges it ⚖️. It defines alignment with reality. That’s why it must be developed, not just labeled. And here’s where I see parallels with software engineering: 1. 🔍 Observability = Continuous Curation Software systems rely on observability and SLAs to stay reliable. Inputs shift, environments change, and so must the system. 👉 Ground truth must evolve too. GenAI output changes with prompts, models, and context. You need to monitor drift, sample new data, and update your ground truth accordingly. 2. 🧬 Source Code = Canonical Ground Truth Source code defines software behavior. 👉 Ground truth plays the same role in AI. It defines what “correct” means. If it’s vague or inconsistent, models behave erratically. Ground truth must be authored, versioned, and owned – just like code. It’s a cross-functional effort across product, research, and labeling teams. 3. ⚙️ CI/CD = Labeling Pipelines Software ships through pipelines. 👉 Ground truth should too. Curation, labeling, and validation must be operationalized. That ensures speed, consistency, and confidence. 4. 🧑‍⚖️ Code Review = Ground Truth QA Code reviews catch bugs and surface disagreements. 👉 Ground truth needs review cycles too. Disagreements, especially in subjective domains are signals. You need structured QA, conflict resolution, and expert oversight. Expert-in-the-loop systems matter here. 5. 🗃️ Version Control = Ground Truth Snapshots You can’t compare models or prompts without knowing the exact ground truth used. 👉 Ground truth must be versioned: immutable snapshots, diffs, and lineage, just like code. Without it, evaluation isn’t reproducible or trustworthy. So yes, ground truth must be developed with the same rigor as code. It’s what your AI is judged against. If you wouldn’t ship code without tests and reviews, you shouldn’t ship AI without the same for ground truth.

20

How do you know when a shortcut is smart? A recent convo with Brittian hit me at just the right moment and drove me home. We are building something Stellar, and were deep in the weeds of a database design. But what made the discussion valuable wasn’t just the technical guidance, it was the intentionality behind it. I had started off designing for scale, thinking long-term and high volume. Naturally leaning towards building it "right". But as we discussed more, the clearer it became: - This use case? It’s niche. - A one-off. - Not something that needs to stretch or scale. That changed everything. It reminded me of that scene from BlackBerry I reflected on last week: “Perfect is the enemy of good.” “Good enough is the enemy of humanity.” This decision sat right between those two ideas. Over-engineering would’ve been the real risk here. Because context is the architecture. The best engineering decisions aren’t about technical ideals. They’re about understanding the lifecycle of what you're building, and why. Not all shortcuts are created equal. Some are just… well-placed bets. #tradeoffs #engineeringdecisions #buildwithintent #startupengineering #productthinking #devcansol #stellar

29

4 Comments

Great conversation with the BQE Software Industry Advisory Board yesterday. Some really insightful comments on what those early in their A&E careers -- i.e., the future leaders in the industry -- really care about. ✔️ "Career development, it's a big thing right now. A clear career path tied to compensation, tied to goals that's what this new generation wants. Training and development is now the highest ranked, above other benefits like 401K. They really care about development and learning." ✔️ "Speaking about the difference between the generations, the younger generations are also much more concerned with how businesses operate. I think the younger generation is more hungry to be independent." ✔️ "Kids go to work that there's already a basic framework that's far stronger than it was 30 years ago. And I think again, it kind of goes back to the the the type of software product that you all are producing. Is it something that especially the young generation would like to use? They love stuff like this. They don't see it as a burden. So it might be a bottom up embracing of what you do. We seem to think it's all top down, but it maybe it's bottom up." What generational trends / themes are you seeing?

58

2 Comments

The tech industry's "good times" are over—so now what? Sean Goedecke delivers a candid, insightful breakdown of how the software engineering landscape has changed. The days of lavish perks and unquestioned autonomy are fading. As rising interest rates reshape company priorities, engineers must realign their expectations and understand the new value equation: deliver outcomes that matter to leadership, or risk being left behind. https://lnkd.in/gyCr5rd2 #TechCareers #SoftwareEngineering #EconomicShift

14

1 Comment