About
Dan is an accomplished programmer, teacher, mentor, and leader. He has over 20 years of…
Activity
2K followers
-
Dan Levy reposted thisDan Levy reposted thisLast week my role at MagicSchool AI got eliminated in a company-wide restructuring. Not exactly the surprise party I was hoping for, but I’m proud of what we accomplished and excited for what’s next. In just over two years I helped build an AI-powered EdTech platform that now serves 6M+ educators and students across 7,000+ schools and districts worldwide. We’re saving teachers 7 to 10 hours a week and driving a 28% increase in student engagement in classrooms where it’s deployed. A few things I’m especially proud of from that run: 📊 Delivered a student analytics dashboard now used by 60K+ educators that turns messy usage data into clear insights on adoption, patterns, and outcomes 🛡️ Led a full build of our student-safety moderation pipeline from requirements and architecture through to production 💰 Designed usage limits for the free tier that cut our annual LLM costs by about $150K 🔗 Built the company’s first Learning Management System (LMS) integrations I’m a Senior Full Stack Software Engineer who loves owning problems from start to finish. I take them from early requirements all the way through architecture and into production. Before focusing on engineering, I spent over a decade in product and program leadership roles, which gives me a strong ability to bridge technical execution and business needs. Currently based in Apex, NC and open to Senior Software Engineer (Full Stack or AI focus), Technical Product Manager, or Technical Program Manager roles. Hybrid or remote both work for me. If you know a team that needs someone who can take ownership and drive things forward, I’d really appreciate an intro. Excited to begin the next chapter and bring that same energy and ownership to a new team. Thanks in advance! #OpenToWork #SoftwareEngineer #FullStack #AI #TechnicalProductManager #TechnicalProgramManager #TechHiring #TechJobs
-
Dan Levy reposted thisDan Levy reposted thisyou could’ve found a job by now if you didn’t have to apply to each job twice
-
Dan Levy reposted thisDan Levy reposted this"The Slow" don't exist. They're responding to your incentives. When a project is scoped as "all of all the goals," you've built the perfect breeding ground for slowness, and then blamed the people caught in it. Slash their schedule and they don't get faster; they start hiding progress and inflating estimates. Now nobody knows anything. The missing concept is scope. Not "add a temporal constraint." That just forces lying. Cut scope. How much of which goal, first? Do that and you get sooner, cheaper, and better at the same time, because you've stopped trying to compress the incompressible. And because the landscape keeps shifting, less scope also means more feedback, sooner. Scope is the steering wheel. Most teams grip the schedule and wonder why they can't turn.
-
Dan Levy posted thisMost AI teams do not have an AI quality problem. They have a measurement problem. Every model comes with impressive benchmarks: MMLU, HumanEval, SWE-bench, long-context scores, tool-use demos, whatever leaderboard is hot this week. Useful? Sure. But your product does not run on benchmark questions. It runs on: - your prompts - your documents - your users - your retrieval pipeline - your latency budget - your edge cases - your weird production failures So “which model is best?” is usually the wrong question. The better question is: Which model performs best on our workload, while beating yesterday's cost & performance, with minimal failures? That means evals need to look less like academic scoreboards and more like product engineering: - deterministic checks for things that should never vary - regression cases from real failures - task-specific rubrics - cost and latency tracked beside quality - human review where judgment actually matters - LLM-as-judge used carefully, not magically The benchmarks are not lying. > They are answering someone else’s question. What AI failure has your team still not figured out how to catch before users do?
-
Dan Levy reposted thisDan Levy reposted thisHey gang! I'm looking for my next software role. I spent the last year riding the AI wave, and now I want to take everything I learned and help YOU! I'd love to do any or all of these for your team: * AI-augmented software development grounded in core engineering principles * Player/coach team leadership * AI strategy and coaching * Systems and product design I'd especially like to refocus on building great tools, products, and experiences for people. I'm available for contract and W-2, I'd love to chat about what you're working on!
-
Dan Levy shared thisOne year at MagicSchool AI. What a year! I got to architect platform-level AI systems at the fastest-growing edtech company ever, shipping key platform security features, chat memory and RAG systems, STEM visualization, generative UI, and more. The scale of impact on teachers and students is real, and rare. I remain invested in your success, cheering from the sidelines, and soon as a (very) minor shareholder. For me, the next chapter starts now. I'm looking for senior IC or staff+ engineering roles at companies building foundational AI infrastructure, LLM model security, customized AI harness implementation, or applied AI platforms. If that sounds like a fit, let's connect! DM me or check out my work at my blog https://danlevy.net .DanLevy.net | Come for the code, stay for the cats.DanLevy.net | Come for the code, stay for the cats.
-
Dan Levy shared thisEveryone's "model-agnostic"–until someone asks to switch providers. Suddenly it's not a switch—it's a tour of mismatched options, different auth rituals, ad-hoc cache config, etc. Multiply that by a few teams and repos and you get a configuration surface area that's… not great. Changing provider+model should be boring. Like swapping database URIs: change a string, and you're done. `llm://{provider}/{model}?temp=0.7&max_tokens=1500` The idea: LLM connection strings—a standard URI format that encodes provider + model + runtime parameters, giving you one parsing/validation layer across stacks and environments. Check out my write-up + examples: https://lnkd.in/gyuAS5Jd Platform & engineering folks: would you standardize on URIs like this, or do you prefer manifests (YAML/JSON) for model config—and why? #PlatformEngineering #AIInfrastructure #SoftwareArchitecture
-
Dan Levy reposted thisI told you... 2026 is going to be wilder 💜 #MagicSchool #AIOSforSchoolsDan Levy reposted thisWe wrapped 2025 by hosting Build Week at our Denver office, and we’re heading into the new year more inspired than ever. ✨ Coming together in person with Anthropic and Amazon Web Services (AWS) ignited our creativity and helped us strengthen the foundation behind our AI Operating System for Schools. Throughout the week, our work centered on how our next chapter can support educators and unlock even more student potential. When schools ask us what it really means to bring AI into classrooms responsibly, these collaborative experiences shape so much of our answer: shared standards, thoughtful, hands-on collaboration, and partners who care just as much about safety and alignment as they do about innovation. We’re beyond proud of what our teams are building and excited about where we’re headed next! In the meantime, come behind the scenes with us for a sneak peek. Special thanks to our collaborators who made Build Week possible: CeCe Mamallo, Russ LeWinter, PMP, CSPO, Addie Rudy, Bill Magee, Jim Kim , Kevin Hakanson, Gowtham Shankar, Nate Ober, Andie Cockerill, William Hu, Caroline Humphrey, John Cronk, Austin Howe, Dan Levy,Todd Tobin, Jason Donahue, Thomas Landes, Matthew Boeckman, Kate Loyd
-
Dan Levy reposted thisDan Levy reposted this🚨 Update! The CSTA New England Regional Conference 2025 has a new date and format! 🗓️ Now happening virtually on Saturday, December 6th via RingCentral! 💻 No travel, no commute—just a full day of connection, inspiration, and CS learning from anywhere. If you saw earlier posts about the in-person event at Rhode Island College, take note of this change! 👉 I’ll still be presenting on a panel alongside Vanessa L. Miller Shawn Rubin on AI in K-12. The esteemed Mike Grandinetti is the key note speaker! Our discussion will be centered around Balancing Innovation, Student Needs, and Equitable Implementation. I can’t wait to share all the work we've been doing at Skillrise Education with you—now online! 📖 Info: bit.ly/CSTANE2025 📝 Register: bit.ly/CSTANEReg Let’s make this virtual gathering the most engaging yet! 🚀 #CSTANE2025 #ComputerScience #EdTech #CSForAll #VirtualConference #Aieducation
-
Dan Levy liked thisDan Levy liked thisI'm out here just doing side missions at this point.
-
Dan Levy liked thisDan Levy liked thisif you're thinking about starting a company, you should read this everyone thinks that things get easy when your startup reaches PMF, has lots of employees, raises lots of capital, etc. but its actually the exact opposite. the bigger you get, the higher the pressure. I know Series B founders and public company execs that are nervous wrecks. in many ways, there's no way around this. but one series A I know seems to have figured it out. he's happy, healthy, and his company is doing well but no better or worse than others. here's how he does it: - don't eat crap - hit the gym 3x a week and lift heavy - take as many walking meetings as possible - no cell phones in the bedroom - sleep mask and weighted blanket - multi-vitamin, fish oil, magnesium, and lions mane - standing lunch with founder friends every weekend (to vent and hang out) - blocks his mornings for deep work - did a lot of self-help, meditation, etc workshops in the past that he uses tools from it sounds like a lot, but if you do this your life will be 100x better. I've done most and it has worked for me.
-
Dan Levy liked thisDan Levy liked thisBetter Auth is joining Vercel. Bereket Engida and his team built the auth library that developers have made their own, with 4.6M+ weekly npm downloads and more than 850 contributors. Welcome, Bereket and team! https://lnkd.in/gmt4xBDzVercel acquires Better Auth to accelerate open source authVercel acquires Better Auth to accelerate open source auth
-
Dan Levy liked thisDan Levy liked this$200 Billion of GPU capacity is being wasted. Brian and I quit Citadel and Bain & Company to fix it. We believe inefficiency is a software problem. Virtualization is the answer. Virtualization lets many workloads efficiently share the same hardware. For example, with virtualization a single CPU can run hundreds of workloads and a hard drive has every bit filled with workloads spread across the world. Today, every type of computer hardware is virtualized, besides GPUs. CPUs, storage, and memory are all allocated through virtual abstractions. By contrast, a GPU runs a single workload at a time. To quote a cloud exec: “once a GPU is rented, we consider it utilized. It could be sitting idle for months and we can’t use that capacity for anything else.” Across the industry it is estimated that 80% of the $250 Billion of deployed GPU capacity is wasted. Thunder Compute is the first company to solve GPU virtualization. We spent the last four years researching and developing software to allocate GPUs exactly when they’re needed, creating more capacity by eliminating waste. While the whole world is focused on building more capacity, there is a massive opportunity to better utilize the capacity that already exists.
-
Dan Levy liked thisDan Levy liked thisWe just shipped a llms.txt at Node.js Now Claude Code, Codex etc can pull straight from the official Node.js docs, instead of a blog post from three major versions ago. Fewer hallucinated APIs. Fewer deprecated patterns. Just code that matches the runtime you're actually running. https://lnkd.in/edK7viC6
-
Dan Levy liked thisDan Levy liked thisEpic Games built Lore, a version control system designed for game development's mix of source code and massive binaries like 3D models and textures. Unlike Git with Git LFS or Perforce requiring live servers, Lore works fully offline, stores identical data once, chunks multi-gigabyte files for partial uploads, and pulls files on demand. Built in Rust, MIT licensed, with SDKs for JavaScript, Python, C#, and Go, Lore is free and open source, addressing gaps in existing tools. via @twtayaan
-
Dan Levy liked thisDan Levy liked thisThe latest Common Crawl Foundation crawl indexed 20.9 million PDFs plus calendars, RSS feeds, BibTeX, markdown etc. A lot more than HTML! I found that with one SQL query via S3 and Buckets! Common Crawl now publishes its full URL index — 2.1 billion rows — as a Hugging Face Bucket with an S3-compatible API. So I didn't download anything: I pointed DuckDB straight at the bucket and queried all 2.1B rows in place. Nothing downloaded, $0 to read. This is what the new S3 API enables. Any tool that speaks S3 (DuckDB, Polars, boto3 etc.) can query large open datasets where they live. A 2.1-billion-row web index, queried from a laptop in a few minutes, for $0. And for anyone working on document AI: 20.9M PDFs in a single monthly crawl is a good reminder that a lot of the "web" is actually documents.
-
Dan Levy liked thiscredit where credit's due, i'm fairly sure that was a George Miranda gem. 🤓 and did you know the second edition just dropped? honeycomb.io/book !! physical books coming in T minus two weeks!Dan Levy liked thisHad a little internet outage and I was like let’s pick *my favourite* book to read - literally had a “throw your head back, laughing” moment :D “The rarest naturally occurring element on Earth is Astatine (atomic number 85 on the periodic table), with less than 1 gram present deep within the Earth’s crust at any given time. The second rarest element is unallocated engineering time” - The Legendary Charity Majors - Observability Engineering The best book on Observability Engineering by far #Observability #O11y #Engineering
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
-
-
-
-
Publications
-
3 Reasons Promises Beat Async/Await
NodeJS Meetup @ Denver
See publicationLearn 3 key reasons Promises beat other patterns for modern async designs!
Do you feel like JS Promises are broken? Hard to debug? Replaced by async/await? Maybe even hate them? This talk is for you! -
Talk: Heroic Refactoring
React Denver
See publicationDiscussion on the virtues of tiny functions & how they solve problems in the refactoring process.
-
Video: Functional Composition Patterns in JavaScript
See publicationIntroducing a composition library for JavaScript: FUNCTIONAL PROMISES.
-
Review & edits for "ES6 for Humans"
Apress
See publicationWorked with Lead Author Deepak Grover to correct a couple technical descriptions & provided a number of improved code samples.
https://github.com/metagrover/ES6-for-humans
Projects
-
Docker - Project Contributor
- Present
See projectAdded better support for debian flavored OSs. If you've run the docker installer lately, you've run a tiny bit of my code!
-
Open Source: Mongoose - NodeJS/JavaScript ORM for MongoDB
See projectWorking with project leaders via github.com, I discussed, identified and implemented desired API changes on the core classes `Model` and `Query`.
Honors & Awards
-
#148 in Machine Learning category
HackerRank.com
Test Scores
-
ACT
Score: 36
Recommendations received
10 people have recommended Dan
Join now to viewOther similar profiles
Explore more posts
-
Form.io
⏱️ When "faster" means saving years, not months. Vasion cut over two years off their development timeline by standardizing forms and workflows with Form.io. Streamlined development. Improved scalability. Way faster delivery. Here’s how they did it 👇 https://lnkd.in/gZkpMr3n #enterprisesoftware #softwaredevelopment #digitaltransformation
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content