Dmitry Kan

https://lnkd.in/d2m5qc4n After two months of intense use of agentic tools and a sustained period of high-flow programming, I am proud to announce that LinguaKin has officially transitioned into production. As a programmer who has spent years exploring diverse linguistic landscapes—from my current deep dives into Polish, Czech, and Hungarian to my previous work with Germanic and Romance branches—I have often felt that modern learning tools prioritize the wrong metrics. This project was born from a need to escape the "digital casino" model of education, where gamification and constant notifications serve to fragment our attention rather than foster deep comprehension. The foundation of LinguaKin is built upon the "Slow-Tech" movement. In an era where apps compete for every second of your screen time, I have designed a tool that respects your cognitive space. This is an offline-first, distraction-free environment optimized for the serious polyglot who values focused, self-paced study over superficial rewards. By removing the anxiety of "streaks" and the noise of unnecessary notifications, LinguaKin allows the learner to return to the core of linguistics: understanding the shared DNA of language families. My own journey through the Nordic and Germanic branches illustrates the power of this comparative approach. Establishing a base in Norwegian served as the key to unlocking Swedish and Danish through pattern recognition. Similarly, German provided the essential scaffolding for a rapid transition into Dutch and Flemish. Today, I am finding that my simultaneous study of Czech and Polish is providing a direct, intuitive bridge to the broader Slavic family. LinguaKin codifies this methodology, allowing users to visualize morphological and phonological alignments across 19 different families and 73 languages simultaneously. Integrating AI agents into my development workflow was the secret to moving this project from a complex concept to a published reality in record time. This initial version represents a significant milestone in my quest to provide a purely functional, utility-driven reference for the polyglot community. It is a tool for those who believe that learning a language is a profound intellectual journey that deserves a quiet, dedicated space. Whether you are mapping the roots of the Indo-Aryan family or exploring the unique structures of Kartvelian languages, LinguaKin is here to support your growth without the distractions. Stop playing languages. Start understanding them. Technical Deep Dive on Dev.to: https://lnkd.in/d9BrqjJV Project Overview and Manifesto: https://lnkd.in/dHUV7zTm Download on Google Play: https://lnkd.in/d2m5qc4n #Polyglot #SlowTech #BuildingInPublic #Linguistics #TypeScript #Linguakin #LanguageLearning #CleanCode #AgenticWorkflows

10

🎬 At Psyloom, we’ve been testing our Psyche Layer — a psychological control layer for virtual actors. It turns a scene request into an actor’s inner map: intent, conflict, timing, and subtle behavioral shifts that make performances feel coherent and alive. In this short demo scene “Miranda”, you can see how the same setup comes to life when driven by the Psyche Layer. The result: direction lands faster, edits shrink, and the character holds emotional continuity across takes. Read the full breakdown here 👉 https://lnkd.in/dRNWNSCc #AIacting #VirtualProduction #DigitalHumans #Filmmaking #Psyloom

2

2 kommenttia

🔥 New in Episode 3 of Stacked Sessions: Most teams using AI are measuring the wrong things entirely. They're tracking code velocity while missing what's actually happening to their value delivery. Danny Presten reveals why traditional value stream maps became obsolete the moment AI entered your codebase - and shares the blueprint his teams are using right now. 🎙️ Listen here: https://hubs.ly/Q03VLWwp0 What metric are you tracking that might be lying to you? #EngineeringLeadership #ValueStreams #AI #StackedSessions

13

1 kommentti

Valentine’s mode doesn’t end for Euno and Cyera, with Euno now officially live as part of Cyera’s integrations 🏹 A few words on why this makes a big difference for both security and data teams: DSPMs excel at classifying sensitive data in the warehouse. But BI platforms like Tableau and Power BI often connect through shared accounts. As a result, security and privacy teams are left in the dark about who on the business side has access to sensitive data through dashboards, reports, and now AI agents too. Euno closes that gap by extending Cyera’s capabilities into the actual business layer. We propagate DSPM classifications (e.g., PII & PCI) from warehouse columns to BI fields, reports, and dashboards. That means you get: → Full visibility into exactly who has access to sensitive data, and who has viewed it → Proactive alerts you can act on when privacy or security policies are breached, all the way to automatically revoking access when needed I’m confident we’ll keep seeing the lines blur between what used to be separate domains: security and data governance. I’m especially excited about this partnership with Cyera as we help security and data teams prevent sensitive data access, better together.

110

10 kommenttia

608 ms to format 48k files. Treefmt was adopted by Nixpkgs! When we started writing treefmt 3 years ago, our benchmark was always how fast it would run on nixpkgs. And now it’s used nixpkgs to keep all the nix code formatted. Treefmt is open source and can be used in your own monorepo. Thanks @brianmacgee @jflry, @infinisil, …. for helping make this happen. To read more, check the project here: https://lnkd.in/erNiajRH

122

8 kommenttia

Oslo-based Riff has raised $16 million in a Series A round for its business-ready "vibe coding" platform. Riff (formerly Databutton) helps companies self-launch high-impact AI applications and agents. The platform goes beyond typical AI coding tools by providing native data connections, production-grade infrastructure, proven templates, and starter apps. This enables practitioners at all levels to build complex and secure AI solutions that drive real business impact and productivity. The round was led by Northzone, with backing from existing investors Skyfall Ventures, Maki.vc, Sondo Capital, Global Founders Capital, and new investor Illusian. The funding will be used to expand Riff's reach, empowering thousands more professionals to build and deploy impactful AI solutions. Congratulations to the Riff team. Tech.eu: https://lnkd.in/d2SSUY6c Deals like this are reshaping the AI landscape. For weekly analysis on the funding, valuations, and strategic moves that matter, subscribe to the Charted Growth newsletter: chartedgrowth.beehiiv.com #AI #SoftwareDevelopment #VibeCoding #DeveloperTools #LowCode #Funding #SeriesA

1

Your CI pipeline takes 35 minutes. The actual tests? Only 20. What's eating the other 15 minutes? One test. One webhook. One painful wait that blocks your entire team every single run. You submit a quote. A third-party system processes it. The webhook arrives... eventually. 15 minutes later. Meanwhile, 280 other tests sit idle, waiting their turn. Your devs stop trusting the pipeline. PRs stack up. Sprint velocity tanks. You can't skip the test. The wait is real. But you can stop letting it hold everything hostage. The fix: trigger the webhook in global setup, let it cook in the background, then validate after your suite finishes. (Check Below 👇) Now your 280 tests run in parallel while the webhook processes behind the scenes. By the time validation runs, the response is already waiting. 35 minutes becomes 20. Same coverage. Zero compromises. #Playwright #TestAutomation #CI #QA #E2ETesting

6

🚀 I’m excited to share that AI Mode is now available in Estonian, Latvian and Lithuanian languages! Our most powerful search experience yet, AI Mode enables you to perform more complex searches and unearth information in a way that’s more intuitive. 💡AI Mode uses our advanced Gemini model to help you ask nuanced questions that would have previously required multiple searches. 💡 For example, you can now ask questions like: “I'm looking to understand different coffee brewing methods. Make a table comparing the differences in taste, ease of use and equipment needed.” And then continue the conversation with: “and what do I need to know about the coffee bean types”. Give it a try at www.google.com/ai or look for a dedicated button in your search window!

129

3 kommenttia

𝗧𝗵𝗲 𝟯 𝗞𝗲𝘆 𝗢𝗽𝘁𝗶𝗺𝗶𝘇𝗮𝘁𝗶𝗼𝗻𝘀 𝗧𝗵𝗮𝘁 𝗖𝘂𝘁 𝘁𝗵𝗲 𝗖𝗼𝘀𝘁 𝗼𝗳 𝗔𝗜 𝗪𝗲𝗮𝘁𝗵𝗲𝗿 𝗙𝗼𝗿𝗲𝗰𝗮𝘀𝘁𝘀 𝗯𝘆 𝟵𝟬% Last month we held an oversubscribed webinar describing how Earthmover helped customers cut the cost of an AI-weather-model inference pipeline by 90%. We've now turned this into a short blog post summarizing our key findings. (Hint: it's all about I/O optimization.) Post link in comments. If you're doing AI for weather and environmental modeling and are not achieving full GPU utilization, your I/O is probably the culprit. Please reach out to learn how Earthmover can help modernize and optimize these workloads.

64

5 kommenttia

Emergent! https://lnkd.in/euvvB4xi Every day I scroll through LinkedIn and see the same narrative: “AI will replace humans.” “AI agents are taking over.” “Developers will become obsolete.” Let me share something from actual hands-on experience. I’ve been testing AI tools recently — including platforms like Emergent — and honestly? The experience has been… below average. Buggy flows. Unstable stacks. Limited professional readiness. Far from enterprise-grade operations. We are still very far from “AI replacing engineers.” What I’m seeing right now is a market flooded with half-baked products riding the hype wave. Lots of noise. Lots of automation claims. Very little robustness. AI is powerful. AI is transformative. But we are nowhere near replacing structured thinking, system architecture, operational maturity, and human judgment. So fellow DevOps engineers, product builders, and system thinkers — Relax. We’re not being replaced tomorrow. There’s still a long road between flashy demos and production-grade reality. Let’s build better systems instead of panicking about headlines.

1

🚀 Excited to share the June 2025 update to Valyu DeepSearch API! Over the past few weeks we've been hard at work removing every bit of friction from search and retrieval for AI. Today you can tap into three powerful new features without touching a line of code: 🔍 Smarter Intent Recognition DeepSearch now automatically spots Named Entities like authors and titles so your queries like “What is Hongyu's work in novel transformer architecture” get you the exact result you meant. 📄 In-Paper Search Need to pull out a figure or section from an open-access paper? Simply pass the URL and ask (e.g. "results of the paper", "conclusion", "scaling law diagram"). DeepSearch fetches the exact content you need. 🖼️ Web Image Retrieval Whether it’s a diagram, infographic or photo, any images embedded on web pages now come back in the results payload, making it easy to build more visual, context-aware responses. There’s no new SDK to install and no extra flags to set. Works everywhere you build: Vercel AI SDK, LangChain, MCP, you name it, we've got you covered If you're building AI applications that need access to content/search, take it for a spin and let me know what you build! 👉 https://lnkd.in/e2J4cGEk

9

I just learned something cool about Rust, that makes perfect sense. Developers have been wrestling with memory bugs since the invention of programming languages. But, we have higher expectations for the physical world, where mechanical engineering just cannot fail. So, it was interesting to learn how the creator of Rust was inspired by mechanical engineering. Here’s a summary of how it played out: In 2006, Graydon Hoare was a 29-year-old programmer at Mozilla. One day, he came home to his apartment in Vancouver and found the elevator software had crashed again. He lived on the 21st floor. Climbing the stairs, frustrated, he had a simple thought: Why can’t we computer people build an elevator that doesn’t crash? Much of the world’s critical software - elevators, operating systems, infrastructure is written in C or C++. These languages are efficient, but they make it easy to introduce memory bugs that cause crashes and security vulnerabilities. The unreliability of software engineering was infecting mechanical engineering. Hoare opened his laptop and started designing a new programming language - one that could match C in performance without memory bugs. He called it Rust, after a group of fungi he described as “over-engineered for survival.” What started as frustration over a broken elevator has become one of the defining systems languages of today. Now, more companies are moving to Rust. Datadog moved to a Rust-based, real-time timeseries database, Dropbox uses it in their file sync, Cloudflare processes more than 20% of all internet traffic with Rust, Figma uses it for its graphics engine When we started building our data startup, we chose Rust at the core, and we’re loving it.

21

What are Just-In-Time Semantic Models??? My co-founder, Tal Segalov, shared a post now on a new concept Solid is pioneering: Just-In-Time Semantic Models (JIT-SM for short) The idea is simple: 1️⃣ Today, you either have one really big semantic model that you use to help AI access your data, or a lot of smaller ones (as Databricks and Snowflake recommend you do). 2️⃣ Either way, you need to maintain a ton of model information manually, and it gets stale very quickly. 3️⃣ As we announced last week, you can use Solid's platform to generate the model 💪 , but that means there are dozens or hundreds of models constantly maintained in your environment. 💥 So now, Solid can instead generate the semantic model on demand - when it is needed to generate SQL or pull certain data. This way, every request gets its own, highly-curated, model. The model can remain in the cache, or get thrown away after use. No more stale models. No more situations where models don't exactly fit your question, or your question spans multiple models (which isn't supported by the DWH's at the moment). Instead - the model your query needs is generated for it... just, in, time. 🎉 And we're starting with support for Snowflake's semantic models, and its Cortex Analyst capability. Check it out: https://lnkd.in/gJAe-gfq

15

In June 2024 former OpenAI researcher Leopold Aschenbrenner wrote that security of AI models and the list of statistical weights was of the essence - and he warned that foreign power could and would steal these. He advocated that security of these should be treated a matter of national security and not just as company secrets. https://lnkd.in/eqHKGkBN And he was right. Yesterday Anthropic released a report accusing three Chinese AI companies - DeepSeek, Moonshot, and MiniMax - of generating more than 16 million exchanges through approximately 24,000 fraudulent accounts to illicitly extract Claude’s capabilities to improve their own models. https://lnkd.in/eekQTvug This was done violating terms of service and regional access restrictions. Anthropic notes that “illicitly distilled models lack necessary safeguards, creating significant national security risks”. This is fully in-line with Aschenbrenner’s 2024 paper. Anthropic continues: “Anthropic and other US companies build systems that prevent state and non-state actors from using AI to, for example, develop bioweapons or carry out malicious cyber activities. Models built through illicit distillation are unlikely to retain those safeguards, meaning that dangerous capabilities can proliferate with many protections stripped out entirely. Foreign labs that distill American models can then feed these unprotected capabilities into military, intelligence, and surveillance systems—enabling authoritarian governments to deploy frontier AI for offensive cyber operations, disinformation campaigns, and mass surveillance. If distilled models are open-sourced, this risk multiplies as these capabilities spread freely beyond any single government's control.” AI models and statistical weights for the AI and AGI models are matters of national security. We were warned in June 2024 and 1.5 y later we get this report from Anthropic. I foresee this will create a significant urgency on the topic from the US Administration. And we will undoubtedly hear much more about this in the time to come. I also foresee this will impact discussions around tighter restrictions on exporting AI chips and cloud services. Finally, it shows that nations who want to be No 1 in AI need to have the full value chain in-country including chips, hardware, critical minerals, rare earth metals etc. This is just yet one more reminder that we are in an AI race where no one can afford not being No 1. AI will change the World as we know it wrt economy, production, military etc. AI is the Manhatten Project of our time! Therefore Anthropic’s report should be taken very, very seriously.

2

1 kommentti

LangGraph is great for workflow structure. But without trace visibility, support triage can still feel like a black box. We published a practical guide showing how to build a LangGraph support triage workflow with FastAPI and Tokvera, with graph-level and node-level trace visibility so teams can debug decisions, routing, escalation, and summaries with confidence. For teams moving from AI demos to real support operations, this is where things start to get useful. #LangGraph #AIAgents #LLMOps #Observability #PythonDeveloper #FastAPI #SupportEngineering #AIAutomation #Tokvera https://lnkd.in/gFVctWCs

5