https://www.deeper-thoughts-blog.rohithshinoj.com/blog Explore the latest advances in AI systems, deep learning architectures, and machine learning research. In-depth analysis of transformers, attention mechanisms, and neural network interpretability. en-us rohithshinoj@gmail.com (Rohith Shinoj Kumar) rohithshinoj@gmail.com (Rohith Shinoj Kumar) Sun, 08 Feb 2026 00:00:00 GMT https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-moltbook-browsers-vision-for-human-ai-interaction https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-moltbook-browsers-vision-for-human-ai-interaction Todays AI is transactional. We chat, then we close the window. The Moltbook is a conceptual leap forward: a persistent, private, and proactive AI partner that learns from your digital life to assist you. This article explores the vision and the immense challenges of this new HCI paradigm. Sun, 08 Feb 2026 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) HCIAI AgentsPersonal AIPrivacyFuture of AIMoltbook https://www.deeper-thoughts-blog.rohithshinoj.com/blog/beyond-rpa-a-technical-comparison-of-ai-agents-and-process-automation https://www.deeper-thoughts-blog.rohithshinoj.com/blog/beyond-rpa-a-technical-comparison-of-ai-agents-and-process-automation Are AI Agents just glorified RPA bots? This article provides a clear technical comparison between rule-based Robotic Process Automation and goal-oriented AI Agents, breaking down their differences in adaptability, reasoning, and data handling, and exploring how they can work together. Sun, 01 Feb 2026 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI AgentsRPAIntelligent AutomationLLMsProcess Automation https://www.deeper-thoughts-blog.rohithshinoj.com/blog/can-llms-ensure-code-correctness-formal-verification-with-ai https://www.deeper-thoughts-blog.rohithshinoj.com/blog/can-llms-ensure-code-correctness-formal-verification-with-ai LLMs can write code, but is it correct? This article explores the intersection of probabilistic AI and the world of mathematical certainty: Formal Verification. We will examine how LLMs are becoming powerful co-pilots for writing provably correct software and ask if they can ever be trusted to reason on their own. Sun, 25 Jan 2026 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Formal VerificationLLMsAI SafetyCode CorrectnessNeuro-Symbolic AI https://www.deeper-thoughts-blog.rohithshinoj.com/blog/building-evolving-ai-systems-with-continual-learning https://www.deeper-thoughts-blog.rohithshinoj.com/blog/building-evolving-ai-systems-with-continual-learning Humans learn new things without forgetting the old. AI models usually cant. This article dives into Continual Learning, the quest to build evolving AI. We will explore the critical problem of catastrophic forgetting and the architectural, regularization, and replay methods designed to overcome it. Sun, 18 Jan 2026 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Continual LearningLifelong LearningCatastrophic ForgettingAI Plasticity https://www.deeper-thoughts-blog.rohithshinoj.com/blog/federated-learning-for-privacy-preserving-model-training https://www.deeper-thoughts-blog.rohithshinoj.com/blog/federated-learning-for-privacy-preserving-model-training How can we train models on sensitive data that lives on millions of devices without ever collecting it? This article dives into Federated Learning, the decentralized training paradigm that brings the model to the data. We will break down the FedAvg algorithm and explore the challenges of communication, non-IID data, and security. Sun, 11 Jan 2026 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Federated LearningPrivacyDistributed MLAI EthicsFedAvg https://www.deeper-thoughts-blog.rohithshinoj.com/blog/enhancing-language-models-with-knowledge-graphs-and-rag https://www.deeper-thoughts-blog.rohithshinoj.com/blog/enhancing-language-models-with-knowledge-graphs-and-rag Retrieval-Augmented Generation (RAG) grounds LLMs in external data, but what if that data had structure? This article explores Graph-RAG, an advanced architecture combining vector search with Knowledge Graphs to provide more accurate, explainable, and powerful context for LLMs. Sun, 04 Jan 2026 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) RAGKnowledge GraphsGNNLLMsInformation RetrievalExplainable AI https://www.deeper-thoughts-blog.rohithshinoj.com/blog/modeling-complex-systems-with-multi-agent-reinforcement-learning https://www.deeper-thoughts-blog.rohithshinoj.com/blog/modeling-complex-systems-with-multi-agent-reinforcement-learning When multiple AI agents interact, chaos can ensue—or brilliant, emergent strategies can arise. This article explores Multi-Agent Reinforcement Learning (MARL), the challenges of non-stationarity, the power of Centralized Training with Decentralized Execution (CTDE), and how MARL is used to model complex real-world systems. Sun, 28 Dec 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) MARLReinforcement LearningAI AgentsComplex SystemsEmergent Behavior https://www.deeper-thoughts-blog.rohithshinoj.com/blog/data-poisoning-a-critical-threat-to-foundational-ai-models https://www.deeper-thoughts-blog.rohithshinoj.com/blog/data-poisoning-a-critical-threat-to-foundational-ai-models Foundational models are trained on web-scale data, but what if that data is hiding poison? This article explores the mechanics of data poisoning attacks, how they create stealthy backdoors in AI models, and why they pose a critical threat to the entire AI supply chain. Sun, 21 Dec 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI SecurityData PoisoningAdversarial AttacksBackdoorsLLM Security https://www.deeper-thoughts-blog.rohithshinoj.com/blog/efficient-ai-at-the-edge-with-on-device-architectures https://www.deeper-thoughts-blog.rohithshinoj.com/blog/efficient-ai-at-the-edge-with-on-device-architectures Beyond the cloud, the next AI revolution is on-device. This article breaks down the architectural strategies (like MobileNets), compression techniques (quantization, pruning), and compiler magic required to run powerful AI models under the tight constraints of edge devices. Sun, 14 Dec 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Edge AIOn-Device AIMobileNetQuantizationPruningOptimizationTFLite https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-scalable-architecture-of-mixture-of-experts-models https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-scalable-architecture-of-mixture-of-experts-models A deep dive into Mixture-of-Experts (MoE), the architecture powering models like Mixtral. Explore how MoE enables massive parameter counts with a fraction of the computational cost through sparse, conditional computation, and the role of gating networks and load balancing. Sun, 07 Dec 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Mixture of ExpertsMoELLMsScalingDeep LearningArchitecture https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-ai-compiler-bridges-models-and-hardware https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-ai-compiler-bridges-models-and-hardware We celebrate the AI models, but not the unsung heroes that make them run fast. This article dives into the world of AI compilers like TVM and MLIR, explaining how they translate high-level model graphs into optimized code for diverse hardware targets through techniques like operator fusion, tiling, and even ML-driven auto-tuning. Sun, 30 Nov 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI CompilerMLIRTVMDeep LearningOptimizationHardware https://www.deeper-thoughts-blog.rohithshinoj.com/blog/graph-neural-networks https://www.deeper-thoughts-blog.rohithshinoj.com/blog/graph-neural-networks From social networks to molecular structures, much of the worlds most valuable data is not in tables or sequences, but in interconnected graphs. This article dives into how Graph Neural Networks (GNNs) work, what makes them unique, and why they are becoming an indispensable tool in modern AI. Sun, 23 Nov 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) GNNDeep Learningmultimodality https://www.deeper-thoughts-blog.rohithshinoj.com/blog/state-space-models https://www.deeper-thoughts-blog.rohithshinoj.com/blog/state-space-models For years, the Transformer has been the undisputed king of AI architectures. But a new contender has emerged: the State Space Model (SSM). This article dives into the mechanics of SSMs like Mamba, explaining how they combine the strengths of RNNs and CNNs to process sequences with linear-time complexity, potentially revolutionizing long-context reasoning and model efficiency. Sat, 15 Nov 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI ArchitecturesLLMsTransformers https://www.deeper-thoughts-blog.rohithshinoj.com/blog/ai-agents-and-autonomous-systems https://www.deeper-thoughts-blog.rohithshinoj.com/blog/ai-agents-and-autonomous-systems Beyond simple chatbots, the next evolution in AI is the rise of autonomous agents. This article explores the architecture of modern AI agents, breaking down the core components of planning, tool use, and memory that allow them to execute complex, multi-step tasks and move us closer to truly autonomous AI systems. Sat, 08 Nov 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI AgentsAutomationLLMsRAG https://www.deeper-thoughts-blog.rohithshinoj.com/blog/machine-unlearning https://www.deeper-thoughts-blog.rohithshinoj.com/blog/machine-unlearning What happens when you need to remove copyrighted data or a user's private information from a trained AI model? The emerging field of Machine Unlearning provides techniques to make models 'forget' specific data without a costly full retrain. This article explores the methods, challenges, and importance of teaching AI to forget. Fri, 31 Oct 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI SecurityInterpretabilityMLOpsMachine UnlearningPrivacy https://www.deeper-thoughts-blog.rohithshinoj.com/blog/physics-informed-neural-networks https://www.deeper-thoughts-blog.rohithshinoj.com/blog/physics-informed-neural-networks Physics-Informed Neural Networks (PINNs) are a groundbreaking class of models that merge these two worlds, creating neural networks that are constrained by the laws of physics. This article explores how PINNs work, why they are revolutionizing scientific computing, and what their future holds. Thu, 23 Oct 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Generative AIInterpretabilityScientific Computing https://www.deeper-thoughts-blog.rohithshinoj.com/blog/visualizing-model-predictions-attention-maps-gradcam https://www.deeper-thoughts-blog.rohithshinoj.com/blog/visualizing-model-predictions-attention-maps-gradcam Why did my model classify this image as a 'cat'? This article dives deep into two key visualization techniques—Grad-CAM for CNNs and Attention Maps for Transformers—to answer that question. We explore the mechanics of how these maps are made, what influences them, and how they serve as powerful tools for debugging model biases and testing robustness. Thu, 16 Oct 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Computer VisionInterpretabilityViT https://www.deeper-thoughts-blog.rohithshinoj.com/blog/introduction-to-explainable-ai https://www.deeper-thoughts-blog.rohithshinoj.com/blog/introduction-to-explainable-ai Deep learning models are often 'black boxes,' making them difficult to trust. This article introduces the field of Explainable AI (XAI) and provides a comprehensive overview of five key post-hoc explanation methods: Occlusion Analysis, Saliency Maps, Grad-CAM, LIME, and SHAP. We explore how each technique works and its role in making AI more transparent. Wed, 08 Oct 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI SecurityInterpretability https://www.deeper-thoughts-blog.rohithshinoj.com/blog/multimodal-ai-architectures https://www.deeper-thoughts-blog.rohithshinoj.com/blog/multimodal-ai-architectures Modern AI is moving beyond single-purpose models to systems that understand text, images, and audio simultaneously. This article explores the core architectural patterns—from early and late fusion to the sophisticated cross-attention mechanisms and joint embedding spaces—that enable multimodal AI to perceive the world in a more holistic way. Tue, 30 Sep 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AttentionComputer VisionGenerative AILLMsMultimodality https://www.deeper-thoughts-blog.rohithshinoj.com/blog/is-your-llm-lying-to-you https://www.deeper-thoughts-blog.rohithshinoj.com/blog/is-your-llm-lying-to-you Why even the smartest AI models sometimes prefer to please you rather than tell the truth — and how to fight back. Wed, 24 Sep 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI SecurityLLMsRAG https://www.deeper-thoughts-blog.rohithshinoj.com/blog/h-infinity-filter-enhanced-cnnlstm https://www.deeper-thoughts-blog.rohithshinoj.com/blog/h-infinity-filter-enhanced-cnnlstm In this article, I explore a novel deep learning architecture from a recent publication of mine that integrates the H∞ filter from control theory into a CNN-LSTM framework to enhance robustness against noise and class imbalance in arrhythmia detection from heart sound recordings. Our approach achieves state-of-the-art performance on the PhysioNet CinC Challenge 2016 dataset, demonstrating its potential for scalable and reliable cardiac screening. Tue, 16 Sep 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Audio ProcessingCNNsComputer VisionHealthcareLSTM https://www.deeper-thoughts-blog.rohithshinoj.com/blog/promptlock-the-ai-malware https://www.deeper-thoughts-blog.rohithshinoj.com/blog/promptlock-the-ai-malware ESET Research uncovered PromptLock — the first ransomware powered by a local LLM. Instead of shipping fixed code, it generates malicious scripts on the fly, making every attack unique and harder to detect. Capable of scanning, exfiltrating, and encrypting files across platforms, PromptLock poses a significant threat to organizations. Read more about how it builds and attacks - prompt by prompt Sun, 07 Sep 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI SecurityDeveloper ToolsGenerative AILLMs https://www.deeper-thoughts-blog.rohithshinoj.com/blog/langchain https://www.deeper-thoughts-blog.rohithshinoj.com/blog/langchain Ever wondered how chatbots like customer support assistants instantly pull up the right product manual page or FAQ? From setup to deployment, lets explore how you can transform raw data into a powerful, domain-aware chatbot in just a few lines of code using Langchain and Chroma to build a Retrieval-Augmented Generation (RAG) pipeline. Sat, 30 Aug 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI EfficiencyDeveloper ToolsLLMsMLOpsRAG https://www.deeper-thoughts-blog.rohithshinoj.com/blog/retrieval-augmented-generation https://www.deeper-thoughts-blog.rohithshinoj.com/blog/retrieval-augmented-generation Large language models dont actually “know” everything—they just predict. Retrieval-Augmented Generation (RAG) changes the game by letting them pull in real knowledge, in real time. Curious how? This blog unpacks RAG, its mechanics, and why it might be the future of intelligent AI systems. Thu, 21 Aug 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI EfficiencyGenerative AILLMsOptimizationRAG https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-ai-video-revolution-with-veo3 https://www.deeper-thoughts-blog.rohithshinoj.com/blog/the-ai-video-revolution-with-veo3 AI video generation is transforming the way we create visual content. Models like Veo 3 can turn simple text prompts into lifelike, dynamic videos, opening new possibilities for storytelling, education, and entertainment. But how exactly do these AI models generate realistic motion and visuals from just a few lines of text? Mon, 11 Aug 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) Computer VisionDeveloper ToolsGenerative AIVideo Generation https://www.deeper-thoughts-blog.rohithshinoj.com/blog/how-do-llms-solve-math-equations https://www.deeper-thoughts-blog.rohithshinoj.com/blog/how-do-llms-solve-math-equations Large Language Models like GPT-4 and Gemini werent built as calculators, but can now solve many math problems. They work by recognizing and reproducing patterns, not by “understanding” math in a human sense, so errors and hallucinations still occur. This blog explores how LLMs handle math and the challenges they face. Sun, 03 Aug 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI EfficiencyGenerative AIInterpretabilityLLMsOptimization https://www.deeper-thoughts-blog.rohithshinoj.com/blog/deploying-ai-in-agriculture https://www.deeper-thoughts-blog.rohithshinoj.com/blog/deploying-ai-in-agriculture What if deep learning could empower farmers to diagnose crop diseases in the field using nothing more than a smartphone? This article explores the implementation of our project, built under a joint ICAR–NITK collaboration, and how Semi-Supervised learning, Vision Transformers, and ONNX deployment was used to create a diagnostic model that achieved 95.7% accuracy—ready for real-world use in the field. Sat, 26 Jul 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AutomationComputer VisionMLOpsONNXSemi-Supervised LearningViT https://www.deeper-thoughts-blog.rohithshinoj.com/blog/setting-up-an-mcp-server https://www.deeper-thoughts-blog.rohithshinoj.com/blog/setting-up-an-mcp-server Learn how to set up a lightweight MCP (Model Context Protocol) server to enable context sharing across AI agents like ChatGPT. This article walks through the core concepts and includes practical code examples to help you build context-aware, multi-agent AI systems. Sun, 20 Jul 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI AgentsAutomationDeveloper ToolsGenerative AI https://www.deeper-thoughts-blog.rohithshinoj.com/blog/llm-vulnerabilities https://www.deeper-thoughts-blog.rohithshinoj.com/blog/llm-vulnerabilities Large Language Models (LLMs) are revolutionizing AI applications across industries. But with their power comes a growing list of security risks—prompt injection, data leakage, and model manipulation. As these threats evolve, one question remains: are we doing enough to protect our AI systems against these threats, and what more can we do? Sun, 13 Jul 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI SecurityDeveloper ToolsLLMsMLOpsOptimization https://www.deeper-thoughts-blog.rohithshinoj.com/blog/visual-perception-through-vision-transformers https://www.deeper-thoughts-blog.rohithshinoj.com/blog/visual-perception-through-vision-transformers Vision Transformers (ViTs) have transformed the way machines process and understand images, replacing convolutions with self-attention. At the heart of this shift lies the question: how do these models “see”? This article explores the internal workings of ViTs, focusing on attention heads, patch embeddings, and interpretability techniques that help us decode how these models perceive the visual world. Sun, 06 Jul 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AttentionComputer VisionGenerative AIInterpretabilityViT https://www.deeper-thoughts-blog.rohithshinoj.com/blog/rise-of-code-llms-and-vibe-coding https://www.deeper-thoughts-blog.rohithshinoj.com/blog/rise-of-code-llms-and-vibe-coding As Cursor transforms how developers write code, we are witnessing the emergence of "Vibe Coding"—a new paradigm where programmers work through intuition and intent rather than explicit syntax. But how are these CodeLLMs different from traditional LLM architectures, and are we really losing the essence of what made us programmers in the first place? Mon, 30 Jun 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) LLMsGenerative AIDeveloper ToolsNLP https://www.deeper-thoughts-blog.rohithshinoj.com/blog/small-language-models https://www.deeper-thoughts-blog.rohithshinoj.com/blog/small-language-models Exploring how 1B-parameter small language models can outperform 405B giants through clever compute allocation, test-time scaling, and innovative training strategies. Is bigger really always better in the AI world? Mon, 23 Jun 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) AI EfficiencyGenerative AILLMsOptimization https://www.deeper-thoughts-blog.rohithshinoj.com/blog/generative-adversarial-networks https://www.deeper-thoughts-blog.rohithshinoj.com/blog/generative-adversarial-networks Ever came across reels on Instagram that show Queen Elizabeth breakdancing and wondered how they look so uncannily real? In a world where AI can create art, faces, and even deepfakes from scratch, GANs stand at the center. But how does AI really generate images that even humans can’t tell are fake? Mon, 16 Jun 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) CNNsComputer VisionGenerative AI https://www.deeper-thoughts-blog.rohithshinoj.com/blog/beyond-attention https://www.deeper-thoughts-blog.rohithshinoj.com/blog/beyond-attention New capabilities of Large Language Models like GPT, LLaMA, Claude, and Grok are emerging almost every day, and at the heart of these powerful models lies the multi-head self-attention mechanism—the secret behind their impressive reasoning and language skills. This article takes a closer look at how these attention heads work together under the hood to create the tools we rely on and admire today. Tue, 10 Jun 2025 00:00:00 GMT rohithshinoj@gmail.com (Rohith Shinoj Kumar) LLMsAttentionNLPTransformersGenerative AIInterpretability