About
Experienced Principal Software Engineer with a focus on front-end development using…
Activity
3K followers
-
Kirby Domingo shared thisShould be a interesting watch, esp with the growth of the gaming/esports industry as a whole!Kirby Domingo shared thisIf you've been considering working in esports/gaming, then you definitely won't want to miss out on this panel. Should have a ton of valuable information so be sure to tune in!
Kirby Domingo commented on a post
5y
Saba Karim would you put some kind of “beta” or “early access” notation? I’m debating this right now. I’ve “launched” a couple times, which essentially is just updating the site/app and further promos via Reddit, ig, twitter. Wondering if it even matters or if people care.
-
Kirby Domingo reacted on thisKirby Domingo reacted on thisLooking forward to joining Glean:GO as a speaker! I’ll be sharing insights on building with the Glean platform using their Web SDK and expanded observability. I’m also looking forward to connecting with other leaders thinking deeply about the future of AI at work. Hope to see you there: https://lnkd.in/gMthEhyJ #GleanGO
-
Kirby Domingo liked thisKirby Domingo liked thisThere’s a good chance your AI is giving you wrong experiment results �� and it’s hard to detect. I’m seeing this pattern across teams building AI into experimentation platforms: 👉 The LLM is doing data work. It feels powerful. It demos well. It’s quietly dangerous. What’s going wrong Teams are: - Writing SQL inside prompts - Letting the LLM pick filters / joins - Computing metrics (even SRM) in the model And it mostly works… Until it doesn’t. The problem isn’t obvious You don’t get errors. You get: - Slightly wrong numbers - Clean-looking charts - Confident explanations The output looks right — but it’s wrong. And now you’re making decisions on top of it. What actually works The Solution: Draw a Hard Boundary 1. The Data Layer (MCP) Role: Deterministic execution. Truth: It fetches data, runs pre-written SQL, and calculates SRM via tested scripts. The LLM cannot change the math. 2. The Orchestration Layer (Skills) Role: Reasoning and synthesis. Truth: It looks at the results from the MCP and explains the "why." The LLM provides the intuition. Simple example “Why is my experiment down on mobile?” - MCP → fetches data + runs SRM - Skill → explains why Takeaway If your LLM is touching raw data logic, you’ve already lost. MCP for correctness. Skills for insight. That boundary is the difference between: 👉 trusted decisions 👉 and silent failures Curious — how are others separating orchestration vs data in your AI stack? #AIEngineering #Experimentation #DataEngineering #LLMOps #SoftwareArchitecture
-
Kirby Domingo liked thisKirby Domingo liked thisCan $9 of API spend and a weekend of prompting beat the betting public? 100M+ tokens, $9 in API costs, and one DeepSeek-powered CLI agent later — here is the post-mortem. Core Takeaway: The single biggest lift didn't come from more data, but from changing the objective function. Instead of asking, "Will this horse win?" (Binary Classification), I switched to LambdaRank (XGBRanker). In a race, you don't need to know if a horse has a 51% chance to win; you need to know where it ranks relative to the field. The Performance (638 non-maiden races): - Major Tracks (AQU, GP, KEE, OP, SA): 44.1% win rate | +35.8% ROI - Minor Tracks (18 tracks): 40.4% win rate | +12.4% ROI Where the Edge Lives: - Turf Sprints: The most profitable cell (+113% ROI on Major tracks). - Large Fields (9+ horses): +68.3% ROI on Major tracks - Claiming Races: The public consistently underrates "class droppers" on Major circuits. - Ranking > Binary: LambdaRank outperformed my previous binary models by +22pp. The Workflow (The "AI User" Part): I didn't write this line-by-line. I orchestrated it through Claude Code + DeepSeek. The agent: - Built the feature engineering pipeline (92 features). - Handled messy PDF parsing and BigQuery queries. - Ran hyperparameter sweeps and backtested a dozen variants. Total R&D cost: $9. A single winning $20 bet at Santa Anita pays for the entire compute budget. The Takeaway: A few years ago, testing a complex pivot like LambdaRank vs. Binary would have been a week-long sprint of refactoring. With Claude Code + DeepSeek, I explored that entire R&D path for $9 in a single afternoon. The real edge isn't just the model; it's the ability to iterate at the speed of thought. #machinelearning #datascience #xgboost #lambdarank #sportsanalytics #ai #deepseek #python
-
Kirby Domingo liked thisKirby Domingo liked this"Dad, do you just take meetings all day?” To be fair, she’s not entirely wrong. But yesterday we cancelled my usual schedule for bring your child to work day! It was great sharing the world of Yahoo Sports and Yahoo Finance with my daughter for Take Your Child to Work Day. Seeing the office through her eyes was a nice reminder of why I enjoy what I do. Huge thanks to Yahoo and the organizers for putting this together. It was a great day! #TakeYourChildToWorkDay #YahooLife #GirlDad #DataScience
-
Kirby Domingo reacted on thisKirby Domingo reacted on thisI am incredibly proud to share that I’ve been promoted to Sr Director, Data Scientist at Yahoo Media Group! 🚀 💪 😃 Reaching this milestone is a beautiful feeling, and it’s one I wouldn’t have achieved without the unwavering support of my family, the guidance of our leadership and the tremendous teammates I have the privilege of working with every day. I am particularly excited about what’s on the horizon as I step into this role: - Multi-Vertical Impact: Expanding my focus to cover both Yahoo Sports and Yahoo Finance. - AI Integration: Leading the effort to weave advanced AI across our entire data world to unlock new insights. - Data-Led Growth: Architecting a comprehensive growth strategy driven entirely by the power of data. There is a unique professional peace that comes with being in an environment where the vision is clear, the leadership has the foresight to act and the results speak for themselves. I’ve never been more energized to continue thriving alongside this team. Let’s get to work! #Yahoo #Promotion #DataScience #AI #YahooSports #YahooFinance #Leadership
-
Kirby Domingo liked thisKirby Domingo liked thisAI will become the ultimate abstraction layer. Just as jQuery abstracted away browser APIs and React abstracted away DOM updates, AI will abstract away manual coding, library implementation, and syntax. The new 10x engineer won't be defined by lines of code written, but by the clarity of the problems they define
Experience & Education
Recommendations received
11 people have recommended Kirby
Join now to viewOther similar profiles
Explore more posts
-
Raanan Avidor
Experimentation isn’t the hard part. Knowing which decisions are reversible—and which aren’t—is. In the latest episode of Pendo.io’s Hard Calls podcast, Trisha Price and Jeff Lash break down how thinking in one-way vs. two-way doors helps product teams move faster without creating long-term drag. Give it a listen! #ProductLeadership #ProductManagement #DigitalAdoption #HardCalls
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content