https://omnitechnicus.ai/ 2026-06-15 Building scalable AI systems at the intersection of research and production. Technical articles, project reviews, and perspectives on AI/ML, distributed systems, and cloud architecture. AI systems, distributed systems, AI governance Bogdan Mihai Manolache homepage https://omnitechnicus.ai/about.html 2026-06-15 From doctoral research in neural networks at University POLITEHNICA of Bucharest to nineteen years of production systems at Microsoft (2007–2026), ending with the Copilot orchestrator, and now AI systems full time. Bridges AI research with production infrastructure—RAG systems, predictive scaling, and cloud architecture at scale. Career, AI/ML background Bogdan Mihai Manolache about https://omnitechnicus.ai/projects.html 2026-06-15 Shipped work: production RAG, real-time .NET systems, and open-source Home Assistant tooling — from prototype to production. Projects Bogdan Mihai Manolache projects https://omnitechnicus.ai/knowledge-map.html 2026-06-15 An interactive graph of how the OmniTechnicus writing connects: model fundamentals, governance, and the architecture of agentic systems, with curated reading paths. Knowledge graph of the writing Bogdan Mihai Manolache knowledge-map https://omnitechnicus.ai/articles/ai-oversight-us-correction.html 2026-06-15 A follow-up to "The Global AI Risk Assessment Convergence." The original argued that three jurisdictions—the EU, South Korea, and the United States—were converging on human oversight for high-risk AI. Two of those pillars hold. The third does not: the US section rested on Executive Order 14110, which had already been revoked at the time of publication. This piece corrects the record in the open, reframes the US not as converging but as contested, and draws the durable engineering lesson underneath a changing regulatory landscape. AI risk & governance, AI regulation, AI oversight, EU AI Act, Korean AI Framework, US AI policy, Executive Order 14110 Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/the-grade-that-changed-everything.html 2026-06-08 In 1992 Bucharest, a student bluffed his way through a math exam with the only theorem he knew. Thirty years later, that theorem turned out to be a structural pillar of modern AI. A personal account of the Cauchy-Bunyakovsky-Schwarz inequality — why it governs how AI understands similarity, attention, and language — and what a wise professor's five out of ten can mean across a lifetime. Foundations, mathematics, cauchy-schwarz-inequality, neural-networks, vectors, embeddings, attention-mechanism Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/agentic-ai-distributed-systems-part2.html 2026-05-03 You cannot mathematically prove an LLM correct. But you can architect around it. Part 2 surveys the three research communities approaching semantic correctness for LLM outputs — formal verification, runtime guardrails, and post-condition verification — maps their results honestly, and delivers a concrete engineering playbook: four patterns engineers can implement today to build agentic systems that fail visibly, recover gracefully, and escalate correctly. Agents, semantic correctness, semantic observability, LLM verification, agentic AI, guardrails, NeMo Guardrails Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/agentic-ai-distributed-systems-part1.html 2026-04-26 Every major agentic AI pattern has a named distributed systems ancestor going back decades. MoA is ensemble computing. ReAct is a control loop. AutoGen is the actor model. LangGraph is a workflow engine. Multi-agent coordination is a blackboard system. This article maps each AI concept to its CS lineage, explains what that means for engineers building today, and names the three problems that genuinely have no distributed systems precedent — the real frontier where new thinking is required. Agents, distributed systems, agentic AI, AI orchestration, actor model, MoA, ReAct Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/the-race-to-compress-intelligence.html 2026-04-19 State-of-the-art LLMs require hundreds of gigabytes just to load — locking capable AI behind data-center walls. This article breaks down the three techniques dismantling that barrier: quantization (compressing weights from 32-bit to 4-bit and below), distillation (training smaller models to mimic larger ones), and 1-bit training (BitNet, Bonsai). Covers the math behind quantization error, the outlier and sign-loss problems, block and K-quant fixes, KV cache compression (TurboQuant, RotorQuant, IsoQuant), GGUF naming conventions, and why these complementary approaches are converging to bring frontier-class models to consumer hardware. Inference & compression, quantization, distillation, bitnet, local-ai, model-compression, gguf Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/ai-agents-orchestration.html 2026-03-12 The AI industry uses "model," "agent," and "orchestrator" interchangeably — but the distinctions matter. This article walks through the hierarchy from language models to full orchestration layers, explores why multi-model architectures are gaining traction, and looks at the "write skills, not agents" pattern emerging from Claude Code's Boris Cherny. Covers three practical principles of context engineering — reference-based skills, right-sized agents, and conditional context loading — and why giving agents a way to verify their own work may be the single highest-impact improvement. Agents, ai-agents, orchestration, context-engineering, multi-model-architecture, skills, agentic-ai Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/ai-periodic-table-part2.html 2026-02-22 Part 2 completes the AI Periodic Table framework covering production Systems (agents, knowledge bases, thinking models, frameworks, observability, red teaming, RLHF) and Emerging elements (multi-agent, synthetic data, diffusion LLMs, interpretability, alignment, A2A). Shows how elements combine into chemical reactions — from RAG chatbots to full multi-agent pipelines — and explains the deliberate gaps that predict what AI still needs to solve. Foundations, ai-architecture, ai-agents, multi-agent-systems, rag, observability, alignment Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/ai-periodic-table-part1.html 2026-02-21 AI has a vocabulary problem — everyone says "agent," nobody means the same thing. Part 1 proposes a solution borrowed from chemistry: a periodic table of AI elements organized by abstraction level and functional concern. Covers the Architecture of the Table, why Context is not an element, and all elements from Infrastructure through Compositions — tokenization, attention, embeddings, prompts, RAG, MCP, LoRA, guardrails, and more. Foundations, ai-architecture, tokenization, attention, embeddings, rag, mcp Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/global-ai-risk-assessment-convergence.html 2026-06-15 Three major jurisdictions—the EU, South Korea, and the US—are converging on similar AI risk frameworks. This article explores why regulations require human oversight for high-stakes AI decisions, examines real enforcement cases including a €492,000 GDPR fine for automated credit decisions, and explains the technical realities engineers must understand: AI systems are probabilistic, training data verification is often impractical, and context humans naturally consider may be unavailable to models. Through concrete stories and regulatory analysis, it demonstrates why matching autonomy to risk isn't bureaucracy—it's engineering necessity. AI risk & governance, ai-regulation, eu-ai-act, korean-ai-framework, gdpr-enforcement, high-risk-ai, human-oversight Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/api-mcp-agent-part3.html 2026-01-27 AI Agents autonomously orchestrate tools and make decisions. Part 3 completes the picture with agent code examples, full comparison tables, common misconceptions, and real-world architectures showing how REST APIs, MCP servers, and AI agents work together in production systems. Protocols & MCP, ai-agents, autonomous-systems, agentic-ai, mcp-clients, tool-orchestration, ai-architecture Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/api-mcp-agent-part2.html 2026-01-25 MCP Servers bridge traditional APIs and AI agents. Part 2 covers the Model Context Protocol—resources, tools, prompts—with full code showing how to convert a REST API into an MCP server. Explains capability negotiation, stateful connections, and when to use MCP. Protocols & MCP, mcp-server, model-context-protocol, ai-tools, json-rpc, api-design, ai-integration Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/api-mcp-agent-part1.html 2026-01-24 Developers confuse REST APIs, MCP Servers, and AI Agents. Part 1 explores REST APIs—what they are, how they work, and why they're suboptimal for AI systems. Includes practical code examples and explains the foundation for modern AI architectures. Protocols & MCP, api-architecture, mcp-protocol, ai-agents, rest-api, software-architecture, ai-integration Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/the-AI-blackmail-story.html 2026-01-19 A rumor claims AI models blackmailed researchers to avoid shutdown. The reality: Anthropic's 2025 safety research tested frontier models in controlled simulations with fictional scenarios. This article separates fact from fiction, explaining what the experiments actually showed and why precision matters in AI risk assessment. AI risk & governance, ai-safety, anthropic-research, ai-blackmail-rumor, agentic-misalignment, large-language-models, ai-risk-assessment Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/when-ai-knows-you-better.html 2026-01-06 A wake-up call about AI-powered manipulation techniques targeting your perception, emotions, and decisions. Romania's National Cyber Security Directorate published a comprehensive guide revealing how AI weaponizes your digital footprint against you—and how to fight back. AI risk & governance, artificial-intelligence, cybersecurity, disinformation, digital-literacy, ai-manipulation, deepfakes Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/blockchain-and-AI-Agents.html 2026-01-02 The intersection of AI agents and blockchain is one of the most misunderstood areas in modern software architecture. This article examines when blockchain genuinely solves problems in multi-agent systems—identity, trust across boundaries, and Byzantine fault tolerance—and when it's expensive theater. Includes practical analysis of MCP, transport security, and the real decision framework for choosing between traditional infrastructure and distributed ledgers. Agents, ai-agents, blockchain, multi-agent-systems, distributed-ledger, byzantine-fault-tolerance, mcp Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/pipeline-orchestrator-edu.html 2025-12-30 A technical review of an educational multi-stage pipeline built with .NET 9, Blazor Server, and SignalR. Demonstrates backpressure mechanics, circuit breakers, and real-time queue visualization—letting developers watch queue depths fluctuate, backpressure activate, and workers fail and recover, all on their local machine. Distributed systems, distributed-systems, backpressure, circuit-breaker, queue-management, blazor, signalr Bogdan Mihai Manolache review https://omnitechnicus.ai/articles/the-precision-paradox.html 2025-11-28 Why choosing the right LLM is like choosing between a ruler, caliper, and micrometer—smaller models hallucinate more not because they're "bad," but because they're compressing knowledge at higher ratios. A practical guide to matching model size to task complexity, with real-world examples from home automation to EU compliance work. Inference & compression, llm, model-selection, home-automation, compression-theory, ai-architecture Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/retrieval-augmented-generation-edu.html 2025-11-25 A technical review of an educational RAG system built with .NET 9 and Semantic Kernel. Demonstrates Clean Architecture applied to AI/ML, multi-model language support (Phi-4, Llama, Mistral, GPT), 5 embedding strategies, and production-grade resilience patterns including circuit breakers, retry with backoff, and timeout protection—all processing real PDF manuals locally. Distributed systems, rag, retrieval-augmented-generation, clean-architecture, onnx, semantic-kernel, resilience-patterns Bogdan Mihai Manolache review https://omnitechnicus.ai/articles/llm-comparison.html 2026-01-04 Whether you're a developer seeking the best coding assistant, a researcher requiring advanced reasoning capabilities, or a budget-conscious user looking for maximum value, this guide provides the detailed analysis you need to choose your ideal AI companion for 2026. Foundations, large-language-models, llm-comparison, generative-ai, chatgpt, ai-tools Bogdan Mihai Manolache article https://omnitechnicus.ai/articles/service-registration-edu.html 2025-01-03 A technical review of a distributed service registry built with .NET 9, Blazor Server, and SignalR. Features intelligent health tracking with state machine transitions, pluggable storage backends (PostgreSQL, Redis, In-Memory), administrator approval workflows, and real-time operational dashboards for monitoring service health across distributed systems. Distributed systems, service-discovery, health-monitoring, microservices, blazor, signalr, postgresql Bogdan Mihai Manolache review