Yousef Alnef

I used to blame the model. Whenever something broke, I’d say “LLMs are unstable. This model isn’t production ready.” It sounded smart. It was wrong. Most LLM failures I’ve seen were infra failures pretending to be model issues. My favorite mistake We added retries after a timeout. One timeout became three retries. Retries hit rate limits. Rate limits triggered more retries. The LLM didn’t hallucinate. Our system panicked. Another classic "LLM is slow today" but Reality A synchronous LLM call hit a tool. The tool hit a DB with no timeout. Everything piled up. Distributed systems 101. Maximum embarrassment. At one point our entire AI feature depended on One model One provider One region We called it an AI platform. It was a demo with a salary. What actually works Strict timeouts and budgets Async queues for long work Caching where safe Circuit breakers that include retries Fallback models or providers End-to-end tracing, not just token counts Strong opinion Models provide intelligence. Infrastructure provides trust. If your LLM system has no backpressure, fallbacks, or degraded mode, users won’t say “infra failed”. They’ll say “AI is unreliable”. Which one hurt you first retries, latency chains, rate limits, or observability? #LLMOps #AIEngineering #ReliabilityEngineering #AWS #Serverless #DistributedSystems #Kubernetes #GDGChandigarh #GDGNewDelhi #GDGAhmedabad #GDGNagpur #rajansharma

3

📡 High-Throughput Microwave Link Test on Motion Platform Here’s a look at our recent point-to-point signal tracking test, using a microwave antenna system integrated with an Aviat modem. The setup consisted of two antennas placed 1 km apart, with one installed on a swing platform to simulate the motion of a vessel at sea 🌊 The test aimed to verify tracking stability, link throughput, and signal recovery under continuous movement. 💡 Test Highlights: • Stable link maintained throughout the full motion range • Throughput exceeded 1.5 Gbps across a 1 km distance • Fast reacquisition during motion reversal The results confirm the system’s capability to deliver high-capacity, reliable microwave connectivity even in dynamic maritime conditions. #KINGSAT #MicrowaveAntenna #PointToPointLink #MicrowaveLink #Aviat #MaritimeConnectivity #AviatNetworks #AntennaTesting #ThroughputTest #MaritimeEngineering #TrackingSystem

123

2 Comments

🔧 Firmware getting bigger? Maybe your language should, too. 🔧 We all love C for its bare-metal control. But once you add an RTOS, networking stack, and safety layers, your source tree can turn into spaghetti. That’s when modern C++ shines. Where C++ Earns Its Keep: ⚡ Zero-Cost Abstractions - Use classes, templates, and RAII. Skip exceptions, RTTI, and the heap. Get safer, cleaner code—no runtime hit. 🏗️ Modularity That Scales - Strong typing and generics = reusable, team-friendly code. Crucial at 50k+ lines. ⏱️ Compile-Time Power - constexpr and TMP push logic to build time, saving memory and cycles—watchdog approved. 🧩 Cleaner Architecture - Encapsulation makes drivers, logic, and config easier to test, certify, and maintain. 🔬 Still Hugging the Hardware - Bit-banging and ISRs? Still easy. Drop into extern "C" for hot paths. The Takeaway Modern C++ helps you manage complexity without losing control. However, with great power comes great responsibility. 💡 If the code keeps growing, maybe the language should, too. #EmbeddedSystems #ModernCpp #Firmware #RTOS #SystemsEngineering

64

50 Comments

💻Networking Ports :- In computer networking, **ports** are logical endpoints for communication. They allow a single device (like a server or a PC) to manage multiple services simultaneously over a single IP address. **Port Numbers** Port numbers range from **0 to 65535** and are divided into three ranges: 0-1023 | **Well-known ports** | 1024-49151 | **Registered ports** | 49152-65535 | **Dynamic/private ports** **Common Well-Known Ports** 20-TCP-FTP (Data Transfer) 21-TCP-FTP (Control) 22-TCP-SSH 23-TCP-Telnet 25-TCP-SMTP (Email sending) 53-UDP/TCP - DNS 67-UDP-DHCP (Server) 68-UDP-DHCP (Client) 80-TCP-HTTP (Web traffic) 110-TCP-POP3 (Email receiving) 143-TCP - IMAP (Email syncing) 443-TCP-HTTPS (Secure Web traffic) 3389-TCP-RDP (Remote Desktop Protocol) ——— ✅Why Ports Matter✅ 〽️ Firewalls:- Control traffic by blocking or allowing specific ports. 〽️ Security:- Open/unused ports can be exploited-this is why port scanning is a common reconnaissance step in cyberattacks.

2

ST Engineering iDirect is partnering with leading Mexican connectivity provider Aitelecom to deploy our next-generation Intuition satellite ground system across Mexico. Together, we're enabling faster network deployment, stronger disaster resilience, and expanded access to reliable digital services, supporting the country’s push toward inclusive digital connectivity. Learn more:

8

Stage 2 — Mapping: When Properties Reveal the System’s Real Problems Mapping is where you stop evaluating components visually and start examining the logic underneath. This means indexing every single property — even the ones that look pointless or poorly named. Those “dumb” properties usually expose the deepest issues. You’ll see patterns like: ➡️ Status + Interaction fused together Buttons with “Info Selected” or “Warning Displayed.” Two different concepts collapsed into one because of past deadlines. ➡️ True/False and Yes/No used instead of booleans “Dark Theme: True/False,” “Has Icon: Yes/No.” Clear signs the system predates proper booleans, variables, or token-driven themes. ➡️ Progress concepts mashed together. “Empty (0%),” “Completed,” “Half Apart/Together.” These should be separate properties — or even separate component variations. ➡️ Font weight as a standalone property on only one or two components A pattern started but never scaled, and something that should live in text variables/tokens instead. ➡️ The classic “Property 1”. Always a red flag for rushed or unfinished logic. To get the most out of this stage, index everything in a tool like Airtable. Table linking lets you cross-reference inconsistencies, duplicate concepts, and naming drift — turning the audit into a clear map of where the system broke down. Tomorrow: Refinement.

1

Understanding the Mobile Core Network – Where Everything Connects Behind every call, message, or data session lies a complex but well-orchestrated system — the Mobile Core Network. This architecture highlights how communication flows seamlessly across multiple layers: 🔹 Access Layer From 2G → 5G, users connect via cell towers, Wi-Fi, and radio technologies. 🔹 Bearer Layer (Transport Network) Ensures reliable data transport through fiber, microwave, and backhaul networks. 🔹 Core Layer • CS Domain: Handles legacy voice (MSC, HLR, SS7 protocols) • PS Domain: Powers data services via IMS, Internet, and applications 🔹 Cloud & Enterprise Layer Supports services like billing, applications, databases, and network management ⸻ 💡 The Real Backbone: Signaling Layer What truly ties everything together is the signaling layer, working across all domains: • SS7 → Traditional voice control • SIP → VoIP & VoLTE session management • Diameter → Authentication, policy, and billing • Roaming & Convergence → Seamless global connectivity across 4G, 5G, and Wi-Fi ⸻ 🚀 As networks evolve toward 5G and beyond, the shift to fully IP-based architectures and cloud-native cores continues to redefine how we design, optimize, and scale telecom systems. For engineers, mastering this end-to-end flow is key to building resilient, scalable, and high-performance networks. ⸻ #Telecommunications #MobileCore #5G #4G #IMS #VoLTE #Networking #CoreNetwork #TelecomEngineering #CloudComputing #SS7 #SIP #Diameter

96

4 Comments

Most AI demos show a single powerful agent solving a task. But real systems rarely work that way. As we move toward building AI Operators, systems start relying on multiple specialized agents working together — detecting anomalies, analyzing context, coordinating actions, and executing remediation. Without proper coordination between agents, things break quickly: ⚙️ Actions overlap 🧠 State drifts across systems 🔎 Debugging becomes a nightmare So the real question becomes: 👉 How do multiple AI agents coordinate reliably in production systems? This week we wrote about exactly that: 🧠 From Single Agents to AI Operators: Why A2A Communication Matters In this article we explore: 🔹 Why single-agent architectures don’t scale to real operations 🔹 What Agent-to-Agent (A2A) communication actually enables 🔹 How task delegation, shared context, and traceable workflows improve reliability 🔹 Architectural patterns for building coordinated AI systems The goal isn't just autonomy. It’s bounded, observable, and deterministic autonomy — the foundation for building real AI Operators ⚙️ 📖 Article: From Single Agents to AI Operators: Why A2A Communication Matters ➡️ Next week we’ll go deeper into the capability layer, exploring how agents interact with tools and systems through MCP (Model Context Protocol). Curious to hear how others are approaching multi-agent coordination in production systems 🤔 #ArtificialIntelligence #GenerativeAI #MachineLearning #AIInnovation #AIOpertors #AgenticAI #AIEngineering #AIArchitecture #MultiAgentSystems 

9

Milestone Reached: 64-bit OS Bootloader & Long Mode Transition I’m excited to share the progress on my custom OS project. I’ve successfully implemented a 64-bit bootloader capable of initializing the CPU into Long Mode. Technical Highlights: Toolchain: x86-64 Assembly (NASM). Memory Mapping: 4-level paging implementation for extended addressing. Security: Configured GDT and enabled the No-Execute (NX) bit. Boot Sequence: Reliable transition from BIOS handoff to 64-bit execution. Priority Roadmap for the Kernel Phase: Phase 1 (Core): IDT setup, Exception handling, and VGA/Serial drivers. Phase 2 (Memory): Page allocation (PMM) and Heap management. Phase 3 (Processes): Context switching and a basic Round-Robin scheduler. Building at this level requires extreme precision where every single byte and bit matters. Looking forward to diving into the Interrupt Descriptor Table next!

1

Right now, in competitive FTTH markets, the incentive structure favors speed. Long-term operational viability tends to surface later. There’s a land grab happening in FTTH. Everyone wants passings. Everyone wants scale. Everyone wants to secure territory before someone else does. So teams get creative. They skip pre-positioning drop infrastructure. They may defer terminal installs. They assume aerial access will resolve as easily as it did in prior regions or projects. They price underground builds from spreadsheets instead of soil conditions. Some of it looks smart in the moment. It accelerates deployment metrics. It protects cost per passing. It keeps capital flowing. But here’s what the market eventually reveals: If you build primarily for near-term metrics instead of long-term operations, the tradeoffs surface over time. Operational shortcuts don’t disappear. They compound. They show up in O&M costs. They show up in customer experience. They show up in municipal relationships. They show up in valuation discussions. Speed matters. Operational durability matters more. The builders who win won’t just be the fastest. They’ll be the ones who built networks someone actually wants to operate and monetize.

69

13 Comments

Building Smarter AI Systems: When Agents Talk to Each Other Today’s AI workflows are evolving beyond single chatbots — they’re becoming multi-agent ecosystems. In this setup, a Parent Agent coordinates multiple Child Agents (like story writers, schedulers, or data fetchers), each handling a specific task. These agents communicate back and forth — sharing data, fixing errors, and learning from feedback — creating a self-improving AI system. This isn’t just automation… it’s collaboration between AIs. Imagine your AI team that talks, corrects, and executes — all without human babysitting. This concept is at the core of next-generation workflow automation — built using n8n + AI Agents. ________________________________________________________________ We share weekly live sessions where we build real AI agents step by step. Subscribe here to stay updated 👉 https://lnkd.in/dBkGp7Ku _________________________________________________________________ #AIAgents #n8n #WorkflowAutomation #ArtificialIntelligence #Automation #AsemaAI #NoCodeAI #AIIntegration #AutonomousAgents

1

I am proud to share that AT&T has significantly enhanced our 5G capacity nationwide, strengthening and accelerating connections for our customers across more than 5,300 cities in 48 states. With the integration of new spectrum, America’s best overall network is now even better, delivering substantial improvements in download speeds and overall capacity. Customers are experiencing up to 80% faster download speeds for mobility and 55% faster speeds with AT&T Internet Air. This advancement is supported by our strategic deployment of EchoStar spectrum through a short-term spectrum manager lease, unlocking meaningful increases in both speed and network capacity. In addition, AT&T has agreed to acquire spectrum licenses from EchoStar Corporation, further reinforcing our commitment to long-term leadership in advanced, high-performance connectivity—across both 5G and fiber. These initiatives position AT&T to continue offering more choices for home internet and wireless services, driving capital-efficient growth and enhancing the overall customer experience. Our ongoing investments in network infrastructure are a testament to our mission: to connect people to greater possibilities wherever they live, work, and play. No one brings wireless and #fiber internet to more places or does it better than AT&T. We are proud to set the standard for advanced networking in America. Your AT&T 5G Is Getting an Upgrade #5G #wireless #connectingchangeseverything

21

In just 12 months, Fort Dodge, Iowa: ✅ Launched a city-owned ISP ✅ Connected thousands of homes and businesses ✅ Created local jobs and reduced resident costs ✅ Built Iowa’s largest municipal broadband network All powered by Nokia’s fiber solutions and Camvio’s cloud-native OSS/BSS platform. A blueprint for what municipal broadband should look like. Learn more about this transformational project → https://lnkd.in/eDphkJSA Know more: https://lnkd.in/euAnhaB5

3

The AI is fundamentally a mathematical model that performs two primary functions: interpolation and extrapolation. Extrapolation can easily go wrong because the model may not be trained in that specific domain/use case, which means caution is needed when applying AI in new areas with insufficient training data. Interpolation, however, is not entirely immune to inaccuracies. For example, interpolating heights by averaging the height of a 1.5 m child and a 1.8 m adult to estimate a teenager's height (1.65 m) is flawed, as growth is not linear. What I mean is that the AI may interpolate different training codebases for a specific use case, but the output could be misleading. The resulting code may appear to be working while hiding security flaws. Where I have personally found AI assistants most valuable is in embellishing my written code, suggesting alternatives, testing, debugging, or enhancing developer tooling. However, the more code I get, the harder it becomes to vet the results, which are often imperfect.

5

5 Comments

The BBU — The True Brain Behind Every Telecom Site Over the years working in the telecom field — from LTE and RAN integration to 5G rollout projects — I’ve seen hundreds of sites with massive antennas, RRUs, and complex fiber networks. But one thing became crystal clear to me: The real power of any site doesn’t come from the hardware you see… it comes from the brain that manages it — the BBU. The Baseband Unit (BBU) is the true thinking engine of the site. It receives and processes every signal, communicates with the MME and the core, and keeps everything synchronized — ensuring users stay connected with the best possible performance. Think of the site as a living system: • Antennas are the arms that transmit and receive. • RRUs are the muscles executing every order. • And the BBU? It’s the brain — analyzing, coordinating, and making the critical decisions that keep everything alive. What I’ve learned is that this isn’t just about telecom systems — it’s about leadership and intelligence. You can have the most advanced setup in the world, but if your BBU (your “brain”) isn’t configured, planned, or managed properly… the whole site fails. And just like that, a strong team or project can collapse without a clear mind and good leadership behind it. Every time I log into a BBU for configuration, integration, or troubleshooting, I see more than parameters and alarms — I see logic, planning, and strategy in action. Each command reminds me that: Intelligence in management matters more than power in execution. The BBU is more than just a device. It’s a philosophy — a reminder that success always starts from the brain before it reaches the antennas. ⸻ #Telecom #BBU #RAN #5G #LTE #NetworkEngineering #Telecommunications #Leadership #EngineeringMindset #SmartNetworks #TelecomProjects

140

10 Comments