Louise Ogilvy’s Post

The 40th edition of the DevNewsletter is out today! 🔗 Here's the link for more details: https://lnkd.in/eZfGTihx Here are the headlines: 🏆 Funding Wins FlexAI Scores £30m Seed Funding to Boost AI Compute Accessibility Octomind Secures $4.8m Investment for AI-Powered Software Testing Solutions Talsec Wins EUR1m Investment for Mobile Security Tools 🚀 Lots of Lovely Launches Revideo (YC S23) Unleashes Open-Source Video Creation Toolkit pgEdge Advances PostgreSQL with DDL Replication and Snowflake Sequences Weights & Biases Launches W&B Weave for Generative AI Deployment LaunchFlow Redefines Cloud Development with Python-Based Infrastructure Tool Last9 Inc's Alert Studio Tackles Alert Fatigue with Advanced Features mabl Introduces Mobile App Testing to Enhance Software Quality 🔦 DevTools of the Week Devtron Inc.: Say Goodbye to Tool Chaos and Hello to Efficiency Neurelo: Cloud API Platform for Your Databases 🗞️ More News Gradle Inc. and GitHub Join Forces to Bolster Software Supply Chain Security SmartBear Boosts API Development with Stoplight Integration GitButler Presents THE MERGE: Where Developer Tools and Communities Converge #DevNews #DevTools #DeveloperTools #Develocity #AI #Collaboration #Cloud

💡Traditional #observability hasn’t delivered on its promise. We need a new approach built on logs, not more versioning or legacy approaches. 🔎 Check out our latest blog to discover why logs matter more than ever. #loganalytics #poweroflogs https://ow.ly/SFkf50TJMJX

Day 5 of #30DaysOfFLCode We have all really cool tools for Federated Learning (FL)—some help train models, others will help with MLOps and Data Analysis as well (like NVIDIA Flare, mentioned in the Day 4 comments by Chester Chen — shoutout to him! I'll keep link to previous post at the bottom). But here’s the thing: will they work in your scenario? And what even are the possible scenarios? "Device" is a tricky term. It could mean: - An MCU with 512 bytes of EEPROM, 8-bit arithmetic (yep, integers are just 8 bits!), 8KB program storage, running on batteries—and there’s a fleet of thousands of them spread across time zones. - Or a beefy server with 64 cores, 64GB DDR5 RAM, a couple of terabytes of SSD, and it’s plugged into the grid. Why does this matter? With an 8-bit MCU on a battery, you really care about resource constraints and optimization. Writing software to fit in 8KB while managing interruptions and counting instruction cycles is practically an art form. (Bad optimization = faster battery drain. And let’s be real, nobody complains about devices losing charge too slowly.) Sure, these cases might sound like TinyML territory, but it’s all part of a spectrum. In many situations, you do need to optimize software for the end device. Different scenarios = different tools for deployment and monitoring. Now, here’s where it gets fun: Check out this (https://lnkd.in/d5E2b8MJ) paper, I mentioned it before. It breaks down FL into three types: - Datacenter distributed learning - Cross-silo federated learning - Cross-device federated learning Each has its own quirks, from the number of devices and connection types to challenges like communication vs. computation, data partitioning, and client statefulness. The point? No one-size-fits-all solution exists. Can we have modular building blocks in form of open standards and/or APIs, to make FL solutions easier to build? My take: Yes. Stay tuned for Day 6, where I’ll dive into this idea :)

Day 4 of #30DaysOfFLCode When it comes to tools in Federated Learning, we usually think about training frameworks like Flower, OpenFL, FATE, PaddleFL. These are great for training your models in a federated way—but they tend to leave out two critical aspects: data analysis and MLOps. That’s where tools like Syftbox by OpenMined come in. Syftbox is a platform designed for developing privacy-enhancing technologies (PETs) applications. The concept is pretty neat: Abstracting a silo as a folder with files shared and synchronised across your network. There are two types of files: - Data files: These contain information you're willing to share with others, like processed statistics enhanced with Differential Privacy (Check out my Day 3 post for more on this). - Executable files: Scripts you can invoke remotely, e.g., for generating new data or performing remote analysis. But will it fit your use case? As my professor used to say, "As engineers, your first answer to such questions should often be: It depends." Are we talking about hospital computer networks as silos or mobile phones? Stay tuned for day 5 to find out :) In the meantime, checkout SyftBox guide (https://lnkd.in/dhn-THge)

🚀 Via's Journey to Optimized Observability! 🚀 💫Key Wins: -Reduced hot storage by 55% -Increased data volume with no extra cost -Adopted by 450+ users 🔎Discover how #Via enhanced efficiency and cut costs with Coralogix! Read the Full Case Study here 👉https://lnkd.in/de8Xzv8R #Observability #TechInnovation #DataManagement #Coralogix #CaseStudy #DevOps

🚀 Excited to see Mitanshu Bhatt leverage Fynix Code Assistant to create fastapi-auth, a robust solution for seamless FastAPI authentication! 💻 Check it out: https://lnkd.in/gb-p4bkt 🌐 Explore Fynix: https://lnkd.in/gRN3-2v8 #FastAPI #Authentication #BuiltWithFynix #AI #Coding #OpenSource #FynixAIAssistant

CodeViz (YC S24) creates a visual map of a codebase to help developers understand and navigate it more easily. Developers usually spend a lot of time reading existing code to gather context instead of writing new code. This is inefficient and frustrating, especially in large or complex codebases. To solve this, CodeViz offers a VS Code extension that generates an interactive map of the codebase, showing everything from high-level system architecture down to specific function calls. This map allows developers to see the actual flow of the code, not just the file structure, which helps them quickly grasp how different parts of the codebase interact. Developers at companies like Amazon, Roblox, and Microsoft use CodeViz to save time and improve their understanding of the code they’re working on. The tool is designed to give all developers the kind of deep context and knowledge that top engineers have, making it easier to write, maintain, and improve software. Founders - Liam Prevelige & Will McCall Check out CodeViz - https://www.codeviz.ai/ P.S - if you need any written content for your early-stage startup... Send me a DM or schedule a call here - https://lnkd.in/d5VeNGkm Ciao👋🏾

🌟 Excited to share my latest project: fastapi-auth – A robust repository designed to simplify authentication with FastAPI! 🚀 It's been an incredible journey bringing this to life, and I'm proud to share that the entire repository was built using fynix.ai Code Assistant (An AI-powered coding tool.) fastapi-auth is designed for developers looking to: ✅ Implement authentication seamlessly in their FastAPI projects ✅ Leverage best practices for security and scalability ✅ Get a jumpstart on building production-ready APIs Building this repository was an enriching experience, showcasing how AI can enhance our development process without compromising creativity or control. 💻 Check out the repo here: https://lnkd.in/gb-p4bkt I'm constantly exploring new ways to innovate and push boundaries in software development. Let me know what you think about the repo – feedback and collaboration are always welcome! 🙌 Thanks to Rohit Negi, aayush mundhra, Ushank Radadiya, Prashant Chiplunkar, Sidharth Nair, Sayli Chaudhari for their incredible efforts in developing such an amazing tool that empowers developers like me to handle end-to-end development with the help of AI. A special mention to Purav Shah and Kushan Shah, the visionaries under whose leadership this tool was brought to life. Your vision and direction have been key to making this innovation possible. 🙌 Fynix AI VSCode Marketplace: https://lnkd.in/gZUQegAC #FastAPI #Authentication #AI #Coding #DeveloperJourney #OpenSource #productivity #devproductivity #FynixAIAssistant #BuiltWithFynix #AIEngineered #MadeWithAI

📣 I've just published the sample repo that uses #AzureSQL, #LangChain and #ChainLit to build an end-to-end chatbot 🤖 application so that you can create solution to chat with your existing data 😍 , with full control over what is sent to the LLMs and how search in the database is done 🤩 (I'm using vector search, but that's just one of the options). In my example I have all details of sessions (title, abstracts, time, speakers, etc.) stored in some tables and I want to have a chatbot to allow you to access that data. 👍 Not even 100 lines of code - white lines included, and it is done, ready to be used! 🚀 #RagHack (Link in the comments)

Introducing 🤗 𝘀𝗺𝗼𝗹𝗮𝗴𝗲𝗻𝘁𝘀, a simple library to build agents by Hugging Face Quick Demo : https://lnkd.in/dTUHH-Ev 🤗 𝘀𝗺𝗼𝗹𝗮𝗴𝗲𝗻𝘁𝘀 - a smol library to build great agents! ✨ 𝗦𝗶𝗺𝗽𝗹𝗶𝗰𝗶𝘁𝘆: the logic for agents fits in ~thousand lines of code (see agents.py in repo shared). We kept abstractions to their minimal shape above raw code! 🧑💻 𝗙𝗶𝗿𝘀𝘁-𝗰𝗹𝗮𝘀𝘀 𝘀𝘂𝗽𝗽𝗼𝗿𝘁 𝗳𝗼𝗿 𝗖𝗼𝗱𝗲 𝗔𝗴𝗲𝗻𝘁𝘀, i.e. agents that write their actions in code (as opposed to "agents being used to write code"). To make it secure, we support executing in sandboxed environments via E2B. ➡️ On top of this 𝗖𝗼𝗱𝗲𝗔𝗴𝗲𝗻𝘁 𝗰𝗹𝗮𝘀𝘀, we still support the standard ToolCallingAgent that writes actions as JSON/text blobs. 🤗 𝗛𝘂𝗯 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻𝘀: you can share and load tools to/from the Hub, and more is to come! 🌐 𝗦𝘂𝗽𝗽𝗼𝗿𝘁 𝗳𝗼𝗿 𝗮𝗻𝘆 𝗟𝗟𝗠: it supports models hosted on the Hub loaded in their transformers version or through our inference API, but also supports models from OpenAI, Anthropic and many others via our LiteLLM integration. ▪️Installation : $𝚙𝚒𝚙 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚜𝚖𝚘𝚕𝚊𝚐𝚎𝚗𝚝𝚜 ▪️Sample Code: -------------------------------------- 𝚏𝚛𝚘𝚖 𝚜𝚖𝚘𝚕𝚊𝚐𝚎𝚗𝚝𝚜 𝚒𝚖𝚙𝚘𝚛𝚝 𝙲𝚘𝚍𝚎𝙰𝚐𝚎𝚗𝚝, 𝙳𝚞𝚌𝚔𝙳𝚞𝚌𝚔𝙶𝚘𝚂𝚎𝚊𝚛𝚌𝚑𝚃𝚘𝚘𝚕, 𝙷𝚏𝙰𝚙𝚒𝙼𝚘𝚍𝚎𝚕 𝚊𝚐𝚎𝚗𝚝 = 𝙲𝚘𝚍𝚎𝙰𝚐𝚎𝚗𝚝(𝚝𝚘𝚘𝚕𝚜=[𝙳𝚞𝚌𝚔𝙳𝚞𝚌𝚔𝙶𝚘𝚂𝚎𝚊𝚛𝚌𝚑𝚃𝚘𝚘𝚕()], 𝚖𝚘𝚍𝚎𝚕=𝙷𝚏𝙰𝚙𝚒𝙼𝚘𝚍𝚎𝚕()) 𝚊𝚐𝚎𝚗𝚝.𝚛𝚞𝚗("𝙷𝚘𝚠 𝚖𝚊𝚗𝚢 𝚜𝚎𝚌𝚘𝚗𝚍𝚜 𝚠𝚘𝚞𝚕𝚍 𝚒𝚝 𝚝𝚊𝚔𝚎 𝚏𝚘𝚛 𝚊 𝚕𝚎𝚘𝚙𝚊𝚛𝚍 𝚊𝚝 𝚏𝚞𝚕𝚕 𝚜𝚙𝚎𝚎𝚍 𝚝𝚘 𝚛𝚞𝚗 𝚝𝚑𝚛𝚘𝚞𝚐𝚑 𝙿𝚘𝚗𝚝 𝚍𝚎𝚜 𝙰𝚛𝚝𝚜?") 💥 𝗪𝗵𝘆 𝗧𝗵𝗶𝘀 𝗠𝗮𝘁𝘁𝗲𝗿𝘀: 🔸Faster: 𝟯𝟬% 𝗳𝗲𝘄𝗲𝗿 𝘀𝘁𝗲𝗽𝘀, crushing industry benchmarks. 🔸Simpler: No clunky abstractions—just code that works. 🔸Versatile: Build smarter agents across any domain. 🎯 Stop waiting. Start building. Try smolagents now and let us know what you think. #LLMs #DataScience #MachineLearning #DeepLearning #RAG #ArtificialIntelligence