Mikhail Isaev

Honored to introduce Core-Ce — the operating system for the Web of Agreement. Core-Ce is a distributed operating system designed for node-based (AKA neighbor) networks where people, infrastructure, and devices coordinate directly through agreements instead of centralized platforms. Core-Ce is open source, enabling developers, communities, and infrastructure builders to contribute, deploy nodes, and expand the ecosystem. Core Specifications Architecture: • Node-native operating system • Web of Agreement protocol • Ce - Block distributed ledger • Autonomous mesh networking • Infrastructure identity system • Decentralized communication layer • IoT and device integration • Open utility layer for applications Core UI Functions: • Mesh Request Broadcast • Node Discovery Panel • Profile Identity Panel • Agreement Feed • Neighborhood Dashboard • Language Control • Node Status Indicators • Broadcast Controls • Direct Node Interaction • Shared Resource Panel • Node Reputation Indicators • Activity Notifications • Agreement Builder • Data Attachments • Infrastructure Control Panels • Network Visualization Core-Ce enables networks where nodes coordinate infrastructure, exchange resources, and establish verifiable agreements across decentralized environments. Sandboxed release---> 5/25/2026 Core-Ce full release. ----> 12/26/2026 #WOA #Ce #ledger #javascript #node #agreementprotocol

7

2 Comments

Tomorrow, theory turns into practice 💥 At the Cursor AI Meetup (Cursor Armenia), one of the key topics will be live demos and practical workflows that highlight how agentic coding works in real development environments. 📌 On February 6, this session will walk through concrete practices teams use when working with coding agents, focusing on real-world application rather than theory. As part of this agenda block, we will cover: 🔹 Best Practices for Coding with Agents and Why Coding Agents Change Development 🔹 How Agents Work (Instructions, Tools, Context) 🔹 Plan First, Code Second 🔹 Context Management Best Practices 🔹 Customizing Agents with Rules and Skills 🔹 Defining Success with Tests and Constraints 🔹 Reviewing and Trusting Agent Output 🔹 Preparing for the Future of Agentic Coding We will also explore advanced setups for larger teams through: 🔹 Cursor Configuration for Big Teams and Big Codebases 🔹 Designing domain-centric codebases for optimal agent context 🔹 Enforcing architecture and requirements to prevent agent drift 🔹 Building small but realistic projects to demonstrate agentic teamwork 🔹 Teaching agents to rely on official documentation for integrations 🔹 Scaling agent usage safely across teams and over time 📆 February 6, 2026 ⏱ 6:30 PM – 10:00 PM 📍 AUA Alex and Marie Manoogian Hall, Yerevan If you’re looking to understand how agentic coding fits into real workflows and real teams, this session will deliver practical insight ✨ Will be waiting for you tomorrow! #AOByte #Cursor #AIMeetup #AICoding #YerevanTech #DeveloperCommunity

10

#TarkLabs #MobileAI #Flutter #OpenAI #TechDevelopment At Tark Labs, we specialize in mobile app development using Flutter. To enter the growing Mobile AI market, we decided to build our own Flutter OpenAI Library. Our goal was to create a tool that would help us stand out and build a strong reputation in the AI industry. We saw developing a library as a way to show our expertise. By creating a useful tool for other developers, we could prove our skills in Mobile AI. This would help us attract clients and grow our presence in this competitive market. We started working on the library with enthusiasm. However, we soon faced challenges. The first issue was that most similar libraries are just wrappers around APIs. These wrappers are simple but not powerful enough for production apps with many users. The second problem was about resources. Building a library takes a lot of time from our senior developers. Since the library itself doesn’t generate revenue, only companies with extra funds can afford this. For us, this was a big challenge. After reviewing these issues, we decided to pause the project. While building a library would boost our reputation, we realized we can’t afford the time and cost right now. We need to focus on projects that bring in revenue to keep growing. In the future, we may revisit this idea when we have more resources. For now, we’ll explore other ways to build our credibility in Mobile AI, like sharing knowledge or smaller projects that align with our goals.

1

Faoliyatim davomida Open Source loyihalaridan ko‘p foydalandim va endi navbat — open-source`ga o‘z hissamizni qo‘shishga. Hamkasbim Otabek Nazarov bilan birgalikda Ovozify Labs loyihasiga start berayotganimizni mamnuniyat bilan e’lon qilamiz. Bizning asosiy maqsadimiz — turli xil ML modellari va dasturiy yechimlarni open source sifatida taqdim etish orqali soha rivojiga turtki berishdir. Bilim ulashilganda ko‘payadi. Bizni qo‘llab-quvvatlaysiz va loyihalarimizda faol ishtirok etasiz degan umiddamiz. --- I have spent most of my career benefiting from the incredible ecosystem of Open Source software. Now, it is time to give back. Together with my colleague Otabek Nazarov , I am thrilled to announce the launch of Ovozify Labs. Our mission is simple but ambitious: to contribute to the global and local tech communities by open-sourcing various Machine Learning models, datasets, and development projects. We believe that innovation grows when knowledge is shared. We are just getting started and would appreciate your support, feedback, and contributions. Check out the full announcement below! 👇 #OpenSource #OvozifyLabs #MachineLearning #AI #Community #Uzbekistan

28

4 Comments

🚀 BIG ANNOUNCEMENT I’m thrilled and proud to share that we've released all the resources and work developed over the past two years at the Tbilisi AI Lab 👇 What’s inside: 🦬 4 powerful models in 3.8B & 12B sizes — pretrained, instruct, and fully aligned. 🦍 20+ datasets: pretraining data, 3M instruction examples, 0.5M DPO pairs, plus function-calling and specific instruct datasets 🐫 Benchmarks: all classical benchmarks translated into Georgian ✨ If you’re working on Georgian language models or small LLMs, this can be a valuable resource for learning, adapting, or improving for your own use case. You can even host it locally and use it however you like - it performs really well ✊ We hope the community will maintain this work and share feedback to help it grow further. 📄 A detailed technical report on how we combined the data and trained the models is coming soon https://lnkd.in/d9xipSFU

111

7 Comments

$10M → $100M → $1B → $9B → $14B. That's one AI company in three years. In 2022, Anthropic made $10 million. A small AI startup building an AI called Claude. By 2023, they hit $100M. By 2024, $1B. By the end of 2025, $9B. And as of February 2026, $14B in annualized revenue. Two years ago, 12 companies paid over $1M/year for their AI. Today, over 500 do. And they're not alone. OpenAI went from $3.7B to $20B in the same period. This isn't about one company. This is about how fast AI is becoming part of how businesses work. Stay curious. That's all it takes. → getrok.com Sources: Sacra, SiliconANGLE, TechCrunch, Foundation Capital, Reuters #AI #Anthropic #OpenAI #AIIntegration #GetRok #Business #ArtificialIntelligence

6

1 Comment

Armenia's $500 Million Nvidia Data Center to Revolutionize AI and Tech in the Caucasus Region Armenia is stepping into the global technology spotlight with a landmark initiative—a state-of-the-art data center built in partnership with Nvidia—poised to reshape both the country’s economic prospects and the wider Caucasus region. Announced by Armenia’s Minister of High-Tech Industry, Mkhitar Hayrapetyan, during the Vanadzor Tech Week, this $500 million investment is more than a technological upgrade; it signals the nation’s aspirations to become a regional hub for artificial intelligence, data processing, and export-driven IT services. Expectations for the project are already high, as forecasts suggest the facility will generate between $200–300 million in annual export revenue—a significant development for Armenia’s growing digital economy. https://lnkd.in/eHiGcypx

5

1 Comment

At Dukakit we have officially begun our journey into Artificial Intelligence. We’ve started by integrating AI to intelligently process and validate M-PESA and other payment transactions. The system now extracts key details such as transaction amount, sender name, and timestamp — even from unstructured messages. This is especially useful for businesses that haven't fully integrated with payment gateways and still rely on SMS-based validation. The result: reduced manual work and improved accuracy across the board. This is just the first step in a broader AI strategy aimed at making business operations smarter, more efficient, and fully data-driven. What’s next for AI at Dukakit? 1. Sales prediction based on historical data and seasonal trends 2. Expense monitoring and anomaly detection for better cost control 3. Wastage and loss tracking across retail, inventory, and production environments 4. Smart invoice generation from voice or text prompts, with contextual pricing suggestions 5. Fraud detection through real-time analysis of irregular financial behavior 6. Employee oversight to flag anomalies in sales, cash handling, or system access 7. Cash flow forecasting, sales trend analysis, and proactive tax readiness alerts Our goal is simple: to transform Dukakit into an intelligent business engine that empowers entrepreneurs with the tools they need to make faster, smarter, and more informed decisions. This is just the beginning of how we’re building the future of business automation. #Dukakit #AI #BusinessIntelligence #SmartFinance #Fintech #SMEAutomation #KenyaTech #SalesForecasting #DigitalTransformation

8

Yesterday, we hosted Corsac’s first webinar with Igor Omelianchuk . For me, it was an opportunity to share how we approach modernization projects in practice. In this post, I’d like to summarize the main points on refactoring, reengineering, and rebuilding, and when each makes sense. 👉 Refactoring: when and why Refactoring is about improving internal code structure without changing behavior. Think of it as reorganizing your office, same people, same tasks, but cleaner desks and smoother workflows. It works when business rules are stable, but the code is messy or difficult to maintain. In one case, replacing outdated libraries and modularizing a .NET Framework app prepared it for migration to .NET Core, all without a code freeze, while still delivering features. In another case, refactoring failed because there were no tests. The team spent more time fixing newly introduced bugs than delivering features. 👉 Reengineering: deeper than code cleanup Reengineering goes beyond code cleanup. It’s needed when architecture itself becomes the bottleneck, for example, migrating from on-prem to cloud, modularizing a monolith, or redesigning for compliance and performance. A good case was a social services project: the team staged the reengineering, moving from a monolith to modular design. Time-to-market accelerated, downtime decreased, and the business gained agility. But we’ve also seen failures. In a healthcare project, management pushed for microservices simply because it sounded modern. With no increase in team size and no preparation for the overhead, deadlines slipped, and costs grew. 👉 Rebuilding: last resort Rebuilding means throwing away the old system and starting from scratch. It’s the riskiest and most expensive option. Projects often take years, budgets expand, and critical business logic gets lost. The business also has to wait until the new system reaches feature parity before seeing value. Rebuilding makes sense only when: The platform is truly unsalvageable or end-of-life. The business wants to completely redesign the product. Or the organization is ready for a long-term reset. In some cases, we recommend hybrid strategies like the strangler fig pattern: gradually replacing legacy modules with modern ones step-by-step while the old system continues to run. That way, you get the benefits of a new stack faster without betting the business on a risky full tech revamp. ⌛ Modernization is not one-size-fits-all. Every system requires a balance between technical feasibility and business goals. Getting that balance right is what makes the difference between wasted budget and a platform that supports growth.

6

The team at DeepFlow have just published "ARCANE: A Multi-Agent Framework for Interpretable and Configurable Alignment," which we'll be presenting at AAAI-2026 in Singapore ARCANE is an elegant and promising approach to bridging a key gap in current LLM-agent systems: the need for interpretable, adaptable alignment in multi-step, tool-using, collaborative workflows. The rubric-based design brings transparency, which is often lacking in black-box RLHF or opaque reward models. Paper link in original post comments.

23

🚨 𝗔𝗻𝗱𝗿𝗼𝗶𝗱 𝗗𝗲𝘃𝘀, 𝗠𝗲𝗲𝘁 𝗬𝗼𝘂𝗿 𝗡𝗲𝘄 𝗙𝗮𝘃𝗼𝗿𝗶𝘁𝗲 𝗔𝗣𝗜 (𝗬𝗼𝘂 𝗝𝘂𝘀𝘁 𝗗𝗼𝗻’𝘁 𝗞𝗻𝗼𝘄 𝗜𝘁 𝗬𝗲𝘁) So I dove into the mysterious depths of Android 16’s android.app.appfunctions API... and I came back alive (with a few gray hairs and one existential crisis). 💡 TL;DR: This API lets your app expose specific functionalities to the system so the OS—and maybe even Gemini—can interact with your app like a socially awkward AI assistant who finally learned boundaries. Picture this: You're building a restaurant app. You expose a function: "Order Butter Chicken" Next thing you know, Gemini’s whispering "Your usual?" to your user like a tech-savvy maître d’. 🔥 Key Facts I Discovered (so you don’t have to): Create an AppFunctionService (because subclassing makes everything fancier) Only system processes can call it (no nosy neighbors allowed) There are permissions like EXECUTE_APP_FUNCTIONS_TRUSTED to keep the creeps out Integration points? Possibly system-wide search and Gemini Use cases? Endless. Automated tasks, voice-triggered features, app search magic... ✨ It’s like Android just gave your app its own VIP pass to the OS. 🤖 Android’s future is starting to think for itself... ...so let’s make sure it thinks your app is worth talking to. 📲 Curious about how to integrate this in your own app without sacrificing your sanity (or build.gradle)? Let’s connect or drop a comment below—happy to chat, debug, or just rant together. #AndroidDev #AppFunctionAPI #Android16 #AndroidDevelopment #JetpackCompose #AndroidEngineers #MobileDev #AppDev #TechHumor #DeveloperLife #GoogleIO #MobileApps #AndroidTips #AndroidHacks #KotlinDev #AndroidStudio #NewAPI #GeminiAI #VoiceInteraction #BoundServices #APIDive #SystemIntegration #MobileInnovation #AIIntegration #TechTrends #DebuggingDiaries

4

We’re excited to welcome Kevork S., Staff ML Engineer at Grid Dynamics and AAAI author, to DevFest Armenia 2025! 🤖 Talk: Operationalizing Evaluations: Framework for Automated CI/CD in the Era of Generative Models for LLMs Traditional testing breaks down in the world of LLMs — outputs are stochastic, non-deterministic, and impossible to validate with classic string-matching tests. Kevork introduces a practical framework for Semantic CI/CD, where LLMs evaluate other LLMs using robust, binary scoring with no humans, no heuristics, and no brittle rules. Building on insights from Gaming the Answer Matcher, he demonstrates how judge models can act as reliable gatekeepers for production pipelines, enabling safer, more predictable deployment of GenAI applications at scale. If you’re working with LLMs, safety, or production systems — this talk is for you. 📅 Dec 20, 2025 | 📍 Woods Center, Yerevan (Jrvezh area) 🔗 More info: devfest.am/2025 #DevFestArmenia2025 #GDGYerevan #GridDynamics #LLM #AI #GenAI #Evaluation #CICD #MachineLearning #ModelSafety

12

🔥 𝗨𝗣𝗗𝗔𝗧𝗘: AI Music Generator Suno in Funding Talks at $2 Billion Valuation Suno Inc., a startup that generates music using artificial intelligence, is in talks to raise over $100 million at a valuation of more than $2 billion, according to people familiar with the matter, quadruple its previous valuation. 🔗 [𝗙𝘂𝗹𝗹 𝗦𝘁𝗼𝗿𝘆](https://lnkd.in/gZyZgjX4) 📢 𝗙𝗿𝗼𝗺 𝗕𝗹𝗼𝗼𝗺𝗯𝗲𝗿𝗴

1

2 Comments

Sarmo Ventures backs Uzbek voice AI startup NavAI with $20K investment NavAI, a startup developing voice AI agents and call center automation tools in the Uzbek language, has raised $20,000 in pre-seed funding from venture studio Sarmo Ventures. The company is building a full-stack platform for speech recognition, voice synthesis, and dialogue management tailored to the Central Asian market. NavAI’s technology enables businesses to automate up to 90% of inbound and outbound calls without human operators. The startup has already launched a demo version of its AI agent, available for testing at navai.uz. https://lnkd.in/d2gy6SBU

20

1 Comment

It seems the SciChart team is knocking out features at a rate of knots ... Today, we have a preview of a Smith Chart in #JavaScript #React Smith charts are used in Electrical & RF Engineering and are a specialised polar plot which represent complex impedances (Z = R + jX). These are used to visualise how impedence changes with frequency and assist in circuit design, antenna design etc... This has been a hotly requested feature for some time, but only now that SciChart has completed its complex polar-plot functionality has it been possible to add it. Coming soon to SciChart.js!

22

1 Comment

HyGPT 1.0 (10b) is finally out! This has been a journey of 6 months, with 3 intensive months dedicated to training. We are proud to present a model that significantly enhances the capabilities of AI for the Armenian-speaking community. Our new model, HyGPT-10b-it, has shown outstanding performance, even outperforming some larger models on specific tasks. Key Highlights: 1) HyGPT-10b-it has demonstrated top-tier results in Armenian, especially in factual accuracy (TruthfulQA) and mathematical reasoning (GSM8K). 2) We've successfully implemented and validated a modification by separating the embedding and output layers, boosting the model's accuracy. 3) Community Collaboration: This project was made possible with the invaluable assistance of Artak Hovsepian (NCCAIT) in data collection and the computational resources from the Nvidia Inception Program. We believe in the power of open-source and have made HyGPT available for everyone. Dive into our comprehensive tech report to learn more about our methodology and findings: https://lnkd.in/eTXku4Ns You can download the model from Hugging Face here: https://lnkd.in/eXAD6V7Y Want to see it in action? Test HyGPT on Telegram: https://lnkd.in/eY7DEm6Q We are excited to see what the community will build with HyGPT!

43

5 Comments

🚀 What a month at Picsart! We shipped four major AI products, and I wanted to share the vision behind them: 1️⃣ Aura — Your AI creative partner The "blank canvas" problem is real. Aura lets you talk your ideas into existence — literally. Voice commands, natural language, seamless image-to-video workflows. Static photo → marketing image → animated video → extended with music. All in one conversation. 60 seconds from idea to trending content. 2️⃣ Persona & Storyline — The faceless economy toolkit The $40B influencer marketing industry is shifting. Brands want consistent characters without talent dependencies. Persona builds digital ambassadors; Storyline keeps their narratives coherent across campaigns. 3️⃣ AI Playground — 90+ models, one prompt Subscription fragmentation is killing creativity. Creators shouldn't need 10 different AI subscriptions. Now access Sora, VEO, Kling, Flux, ElevenLabs and 85+ more through one interface. Pay per generation. No more subscription stacking. 4️⃣ AI Copilot in Flow — From prompt engineer to creative director The best creators aren't tweaking prompts — they're directing. Copilot handles continuity, camera metadata, model selection so you stay in flow. Early users: 3x faster project completion. The common thread? Removing friction between imagination and execution. #AI #CreativeTools #Picsart #GenerativeAI #ContentCreation

129

5 Comments

🚀 In most cases you don’t need any external global state management library. One thing I’ve learned while building React Native apps: we often over-engineer state management by rushing to libraries like Redux, MobX, or Zustand. But here’s the truth 👉 if you follow “One component, one responsibility”, React’s built-in useState and useContext are more than enough for 80% of use cases. 💡 Why? Components should manage only their own state. Share state upwards when multiple components need it (lifting state up). For truly global data (like auth/user), Context + useReducer works beautifully. When you follow this principle, your code becomes: ✅ Simpler to maintain ✅ Easier to debug ✅ Faster to onboard new devs ✅ Less boilerplate External state management libraries are powerful — but they’re like power tools 🛠️. Use them only when you’re working on a project that really needs global, complex state syncing. Start small. Keep it simple. Scale wisely. What’s your go-to approach for managing state in React Native apps? #ReactNative #StateManagement #JavaScript #MobileDevelopment #localstate #useState #globalstate #redux #zustang

2

What if Your Ecosystem Had AI Integrated? Imagine a web-based platform that brings all your installed apps together — where AI can create, edit, and manage files and documents, with a built-in browser for research and much more. ElixirLab Qbit Ecosystem will offer the same seamless functionality, integrated with tools like Google Apps, Notion, and VS Code. via prompting and instructions and multiple mcp servers connected with foundational model , it's like AI controlling the OS #ElixirLabs #ai

7