Liam Ormiston

A couple weeks ago I shared a video explaining embeddings in AI and the response was incredible. Today I'm back with the next piece in the series - how indexing works in vector databases. Indexes are what make vector databases fast enough to power real RAG (Retrieval-Augmented Generation) pipelines in production. If you've ever wondered how we move from “just storing vectors” to reliable, scalable retrieval at low latency, this one's for you.

42

8 Comments

I wrote a blog post about some of the options for adding type checking to your Django project! Historically, Django projects have been tough to type-check, but the future feels bright, especially if PEP 827 gets accepted! https://lnkd.in/eh79eCjV

20

2 Comments

Client-First utility classes in Webflow: things you could didn't notice before 🔥 I’ve been using Client-First for a while, but revisiting its utility classes recently made me realize there are a few subtle features I had completely overlooked. Thought I’d share in case someone else missed them too: 1. Client-First sets "font-smoothing" and "text-rendering: optimizeLegibility" globally This ensures the text on your site looks much closer to what you see in Figma, especially in terms of weight and sharpness. Without these styles, fonts can appear heavier or blurrier in some browsers. 2. Focus-visible styling for accessibility There’s a built-in style for ":focus-visible" on focusable elements like [tabindex] and input[type="file"]. It ensures keyboard users get a clear, consistent focus outline. 3. The ".inherit-color" class is surprisingly useful This class makes all child elements inherit the parent color. Perfect for nested components when you want consistent text color without deep styling overrides. 4. Multi-line truncation with ellipsis Utilities like ".text-style-2lines" and ".text-style-3lines" handle line clamping in a clean, reusable way. Some of these are really simple, but that’s what makes them powerful. Which of these have you used before, or totally missed like I did? #ClientFirst #Webflow #CSSUtilities #FrontendTips #WebDesign

41

2 Comments

A walking-navigation concept using the #Spectacles that redesigns the already great Outdoor Navigation sample to reflect my personal design philosophy. Project your palm downward to create navigation paths on sidewalks, and use your palms to show, expand, or collapse map projections driven by the Snap Places API + Map Component. I think augmenting the activity of walking is extremely interesting, and although this prototype was done in < 2 days, I can already see the possibilities for where navigation augmentation might go next. Thanks to the excellent design teams that built the original Outdoor Navigation project and other AR samples, including Mathieu Vignau Mitchell Kuppersmith Karen Stolzenberg Eric Guo Terek Judi Ilter Canberk Viktoria Hwang Laura Chambers Yuhao Ko Shin Kang Su Hyun (Sue) Lee Alessio Grancini. You all are super inspiring to me! Snap Inc. Snap AR #navigation #spatial #augmentedreality

1,137

61 Comments

Upskilling Chronicles: Week 3, post 1 🚀 🚀 Built A Scraper with Python! I recently challenged myself with a mini project (provided by DevProjects): building a web scraper that pulls article titles, authors, and publication dates, and previews content from news websites and articles. What I Built 🛠️ 🔹 A Python script that scrapes news articles from different sites 🔹 Extracts key details like title, author, date, and content 🔹 Uses BeautifulSoup to parse HTML 🔹 Handles missing data and different website structures 🔹 (Still improving it for sites with bot protection & paywalls!) What I Learned 💡 💻 Web scraping is about pulling text and also understanding how websites are built. 🛑 Many sites block scrapers, so you have to find creative ways to extract data. 🐍 Python’s requests & BeautifulSoup make scraping easy but error handling is key! 💡 A small project like this helped me connect theory to real-world coding. Why Is This Useful? 🤔 This kind of tool has real-world applications like: ✅ Tracking news across multiple sources 📢 ✅ Analyzing content for SEO & trends 📊 ✅ Automating market research 🔍 ✅ Even powering AI models that summarize news! 🤖 Next Steps 🚀 - Handling JavaScript-heavy pages with Selenium - Saving results to a database or CSV - Expanding it to work with more news websites 🔗 Check out the code on GitHub: [in the comments] Live demo video below ⬇️ Would love to hear from fellow devs, have you ever built a scraper? How do you handle sites that block bots? 👀 #Python #WebScraping #LearningByDoing #Automation #BeautifulSoup #SelfTaughtDev #pythonenthusiast

23

1 Comment

"AI sucks at coding. I spend more time debugging then it would have taken to just build it myself" Womp womp. If this is your stance as an engineer you are doing something wrong. You can't just tell the LLM "build this, solve this". You need to provide rich context. You have to teach the LLM how to code like you, by describing how you code This takes time. But it's so worth it.

238

286 Comments

Getting started in software engineering has always been hard, but in 2025 it’s one of the most asked questions. I wrote a couple tips for what I would do in my weekly newsletter. Do - Specialize - Become an AI native technologist - Build a personal brand that accelerates your networking Don't - Try to be Fullstack - Attend a Coding Bootcamp - Wait for your first job to start building something real Have some other tips? Share them below! #TechCareers #SoftwareEngineering https://lnkd.in/e8y4yCUZ

33

11 Comments

AI isn’t coming for jobs – it’s already here. But here’s what most people don’t realize: this is the biggest opportunity rural America has seen in decades. If you know how to build, fix, or create, you’re sitting on a goldmine – and the tools to multiply your skills are free.

11

4 Comments

📍Problem I Want to Solve in My Rural Community While job hunting, I had to move back to my rural hometown to regroup and start fresh. Being back here gave me time to look around -- and I noticed something that stuck with me: Most local schools haven't gone digital. Some have a computer or two. Many still rely on Excel sheets -- or even books -- to manage student records, staff details, and daily operations. There's often no internet or budget for cloud services or monthly software fees. That got me thinking: What if I could build a simple, offline-first system that runs on a local network -- with zero dependency on the internet? 🖥️ So here's what I'm planning to pitch to a nearby school: A LAN-based school management system that works entirely offline. Here's the approach I'm planning to take: 🔧 Setup - One PC acts as the local server(can still be used normally). - Other staff computers connect to it via a local network (LAN). - No internet is needed- everything will sync offline through a router or switch. 🛠️ Tech Stack(Still drafting) - Backend: Node.js + Express - Databases: SQLite or PostgreSQL (Prisma ORM) - Frontend: React - Desktop Wrapper: Electron.js (at this point, it is optional, because of performance issues, it might not be that necessary) - Communication: Server exposes API over static IP, clients will connect to it using Axios/Fetch This isn't built yet -- just the architecture and direction I plan to take if the proposal moves forward. It's personal project, a way to build to build hands-on experience while contributing something useful to the community that raised me. And hopefully, it becomes part of my portfolio too. If you've ever built offline apps or LAN-based systems or have any experience, I'd love your thoughts. Would you change the stack? Improving syncing? Skip Electron? #LearningByBuilding #NodeJS #ReactJS #JobHunting #EdTech

15

4 Comments

Anthropic's Claude had its full system prompt leaked recently - and a lot of people were very surprised to learn that it was ~24,000 tokens long. System prompts like Claude's are meant to allow users who interact primarily through the Web interface to have a consistent and well curated experience. With a number of built in tools, and the desire for consistency across multiple chat windows and many users all trying to get Claude to do all kinds of tasks directly from the browser. Imagine if you unrolled your whole Agentic flow into a single system prompt, including examples for a number of different potential interaction flows - that's why, seemingly, Claude is hogging down tokens in the system prompt. If you'd like to use the model *without* the system prompt - just use the API! https://lnkd.in/gnbx6hcq

124

18 Comments

I'm frequently doing weird stuff in code that doesn't have many (if any) examples out on the internet. I've noticed that LLM code assist tools will do something annoying in these cases, especially if I'm trying to decode something that uses a custom encoding protocol, and its struggling to come up with a solution. It will try to have me make changes to my code which will "fake" a success. If its a decoder it will create a code that produces proper looking output but its not actually decoding anything, just making up cleartext that looks right. I was working with a fairly obscure and not commonly used quantum simulation library, which hasn't been well maintained and wouldn't run. Two different LLMs gave me similar patches which made the library run, but would produce false, totally made up simulation results. If you're doing something challenging, and the LLM is struggling to figure it out, watch carefully to make sure it doesn't provide code that fakes it. Sometimes its quite obvious that's what they are doing, other times they can be quite sneaky.

68

12 Comments

Tailwind CSS just became the most terrifying "AI took my job" story in tech. Not because people stopped using it. Because more people and bots are using it than ever...… and the business still got crushed. This week, Tailwind Labs reportedly let go of 3 out of 4 engineers. Read that again. A tool that's basically infrastructure for modern frontend…... had to shrink because the internet's new behavior isn't "search --> click --> learn --> buy" anymore. It's "ask --> answer --> never visit the source" this is the part that punched me in the gut A developer opened a PR to make Tailwind's docs easier for LLMs to consume. And the founder basically said If we make it easier for AI to read our docs, even fewer humans will show up...… and we'll sell even less….. and the framework becomes harder to sustain. That's the paradox we're all walking into. The open web ran on attention. Open source ran on trust with distribution. And now AI is quietly re-routing both. Stack Overflow didn't "lose relevance" overnight. It got bypassed. Docs didn't "get worse" They just stopped being the destination. So the real headline isn't "AI is replacing jobs" It's this AI is deleting funnels. If your product depends on people reading your content, visiting your docs, clicking your site, discovering your paid tier..… …you’re not competing with another company anymore. You're competing with the new default behavior of the internet And yeah, this is going to get messy. I don't have a neat conclusion. But I do have a question that every builder, creator and SaaS founder needs to answer this year When AI becomes the interface… what becomes your business model? If you build devtools, education, or content-led products, how are you adapting?

14

🚨 Excited to share our latest project: IndySafeBot! 🤖 At the Indy Civic Tech Hackathon 2025, our team built IndySafeBot, an AI-powered public safety assistant that answers emergency-related questions using a RAG (Retrieval-Augmented Generation) pipeline. 🧠 Ask it things like: “What should I do if a volcano erupts?” “How do I report vandalism or graffiti?” “What goes in an emergency kit?” 💬 Built with: Python + Streamlit for the UI FAISS + Sentence Transformers for retrieval FLAN-T5 for answer generation A user feedback loop to improve over time! 📚 The bot learns from corrections and updates its knowledge base—empowering communities to get accurate, real-time safety info. Check out the demo and code: 🔗 GitHub: https://lnkd.in/ddAwJuCX Huge thanks to the organizers and our amazing team! #IndySafeBot #CivicTech #AIForGood #Hackathon #Streamlit #RAG #NLP #PublicSafety #MachineLearning

13

1 Comment

Spent some time yesterday diving into RAG (Retrieval-Augmented Generation) and found these videos to be extremely helpful in understanding how to move from toy projects to production-grade pipelines. If you're working on RAG applications, these are must-watch: 1️⃣ Jerry Liu @ AI Engineer Summit – Talks about challenges in productionizing RAG, evaluation techniques, and how to make retrieval better. Link: https://lnkd.in/gRK8hi2R 2️⃣ LlamaIndex x Arize – Practical tips on fine-tuning embeddings, chunking, and query routing for more robust RAG systems. Link: https://lnkd.in/gdP7RZvW 3️⃣ Langchain RAG Course – End-to-end hands-on tutorial to build a multi-user chatbot using Langchain, FastAPI & Streamlit. Link: https://lnkd.in/gV_e7KGa Definitely worth a watch if you're building with LLMs in the real world.

52

1 Comment