Writing

Aug 10, 2026

SecurityInfrastructure

A SIEM tells you your house is on fire; a SOAR system tries to put it out. Security Orchestration, Automation and Response (SOAR) platforms are the hands and feet of a modern Security Operations Center (SOC). They connect disparate security tools, automate tedious investigation tasks and stop cyber threats in real time without waiting for a human analyst.

Jul 27, 2026

SecurityInfrastructure

Security teams usually drown in data. Every firewall, server and application generates millions of logs daily. A Security Information and Event Management (SIEM) system solves this. It is the central brain of an organization's infrastructure: it collects data from everywhere, normalizes it and uses analytics to find the actual cyber threats hidden in the noise.

Jul 13, 2026

AI SystemsLLMs

Large Language Models predict text well but don't remember facts reliably. When they lack context, they "hallucinate" plausible, fake answers instead. Retrieval-Augmented Generation (RAG) fixes this by forcing the model to read specific, verified documents before it answers, instead of letting it guess.

Jun 29, 2026

AI SystemsAutomation

Early AI integrations required a custom connection for every combination of LLM and tool. Anthropic's Model Context Protocol (MCP) replaces that mess with a single open standard for how AI applications talk to external systems. Pair MCP with an automation platform like n8n and an agent gets access to thousands of tools without custom code for each one.

Jun 15, 2026

AI SystemsAutomation

Building a chatbot no longer requires a decision-tree architecture and months of intent mapping. Microsoft Copilot Studio is one of the tools making it easier to build autonomous agents, moving from rigid conversational trees to agents that reason over a task. By combining Generative Orchestration, Knowledge Sources and Generative Actions, it is possible to create an AI that does not just talk but reasons, retrieves enterprise data and executes tasks.

Jun 1, 2026

AI SystemsLLMs

In my last article, I looked at how Large Language Models (LLMs) predict the next word to generate human-like text. This time I am looking at how that technology transformed chatbots. Older chatbots were rigid decision trees built from if-statements. Today's LLM-driven chatbots act as autonomous agents that understand context, perform multi-step reasoning and turn natural language into the ultimate user interface.

May 18, 2026

AI SystemsLLMs

Large Language Models (LLMs) like ChatGPT, Gemini, Claude and Llama seem like they're thinking, but at their core they are prediction engines. They are built on the Transformer architecture and learn mathematical relationships between billions of words during "training", processing text in units called "tokens". Instead of "knowing" facts, they calculate the most probable next word in a sequence.

May 4, 2026

Privacy

Samarati's algorithm is a solid tool for data privacy, but it applies the same level of generalization to your entire dataset, which can destroy information that didn't need to be touched. The **Mondrian Algorithm** treats data as points in a multidimensional space and recursively slices that space into smaller grids, achieving K-Anonymity with more precision and preserving far more of the underlying data.

Apr 20, 2026

Privacy

In our last article, we talked about K-Anonymity and how it protects privacy by hiding individuals in a crowd. But how do we actually calculate the perfect amount of "blurring" without destroying the data's analytical value? By using Pierangela Samarati's algorithm. Mapping all possible generalizations into a hierarchy and using a binary search, it finds the optimal balance: maximum data utility with guaranteed privacy.

Apr 6, 2026

Privacy

As stated in the previous article, simply removing names and IDs from a database isn't enough to protect privacy. Through "Linking Attacks," supposedly anonymous data can be cross-referenced with public records to re-identify individuals. K-Anonymity solves this by ensuring that every person in a dataset is indistinguishable from at least *k-1* other people. Using techniques like generalization and suppression organizations can release useful data without compromising individual privacy.

Mar 23, 2026

PrivacyMLSecurity

Federated Learning (FL) promises privacy by keeping data on your device, but it isn't bulletproof. Attackers can analyze model updates or outputs to run several types of attacks, including Membership Inference Attacks (MIAs), which try to determine whether a specific individual's data was used to train the model. Adding explainability through Counterfactual Explanations makes these attacks even more effective. During my thesis, I found that adding explanations increases attack success by up to 9%, but applying the right privacy-preserving techniques can cut this risk sharply without destroying accuracy.

Mar 9, 2026

PrivacyML

Removing names and personal information from a dataset is not enough to protect privacy. Modern algorithms can easily re-identify individuals from supposedly "anonymous" data. Differential Privacy (DP) injects a precise amount of noise into the data or model training process, so the presence or absence of any single individual doesn't noticeably change the model's output. DP stops privacy attacks like Membership Inference Attacks, but it comes at a steep cost to model accuracy and that trade-off is what AI engineers have to manage.

Feb 23, 2026

AI SystemsPrivacy

AI models now routinely make life-changing decisions but can't explain why. Explainable AI (XAI) tries to fix this with techniques like SHAP (which features mattered), LIME (local approximations) and Counterfactual Explanations ("change X to get a different outcome"). The EU now legally requires explainability for high-risk AI through the GDPR and the AI Act. But there's a catch: making models more transparent can also make them more vulnerable to privacy attacks. My thesis research showed that counterfactual explanations (one XAI technique) can inadvertently leak sensitive training data.