Iuliia Gorshkova

Software engineering has long embraced the idea that failure is not a question of "if" but "when." We build distributed systems with circuit breakers, database migrations with transaction logs, and deployment pipelines that can revert to a previous state in seconds. Yet, when we transition from deterministic code to probabilistic models, we often leave these [...]

Artificial intelligence has become a driving force for innovation, with startups at the forefront of shaping new solutions for industries ranging from healthcare to finance. As AI systems grow increasingly sophisticated, so do the challenges associated with their integration into real-world workflows. Among these challenges, explainability—the ability to understand and interpret how AI systems arrive [...]

When you're building an AI startup, the allure of the black box is seductive. It promises performance without the headache of understanding the underlying mechanics. You feed data in, magic happens, and insights come out. For a small team racing against the clock and competitors, this seems like a shortcut to a viable product. But [...]

When you’re building an AI startup, the allure of the black box is seductive. It promises shortcuts: just feed it data, and the magic happens. You don’t need to understand the internal mechanics, just the inputs and outputs. For a lean startup trying to move fast, this feels like a competitive advantage. But this approach [...]

When we talk about artificial intelligence in software today, the conversation often drifts toward Large Language Models and generative capabilities. While these advancements are impressive, they represent only a fraction of how AI is being integrated into the world’s infrastructure. The real engineering challenge—and the domain where AI’s impact is most profound and scrutinized—is in [...]

There's a particular kind of silence that falls over a room when a system designed to keep people safe makes a decision that could have, or did, cause harm. It’s not the loud, chaotic silence of an alarm, but a heavy, thinking silence. The kind where engineers in headsets and domain experts with decades of [...]

Machine learning models often feel like black boxes that magically improve with more data. We feed them examples, they adjust their internal weights, and somehow, accuracy creeps upward. This iterative process of trial and error, guided by a feedback loop, is the engine of modern artificial intelligence. Yet, there is a profound distinction between a [...]

There’s a subtle but pervasive myth in modern machine learning: that if you throw enough data at a model and fine-tune it with reinforcement learning from human feedback (RLHF), you eventually get a system that understands the world. It’s a seductive idea because it mimics the way we learn—trial, error, correction. But this analogy breaks [...]

Artificial intelligence has become a cornerstone of innovation, with startups rapidly integrating AI into products to solve an ever-expanding array of real-world problems. Yet amid the race for smarter, faster, and more adaptive systems, crucial engineering considerations are often overlooked. Foremost among these is the issue of memory—both in the computational sense and the broader, [...]

There’s a particular kind of quiet that settles in a server room when a critical model fails silently. It’s not the loud crash of a database outage, but the insidious hum of a system confidently serving wrong answers, hallucinating citations, or optimizing for a metric that no longer aligns with reality. As we integrate these [...]

The Illusion of Self-Validation There is a peculiar irony in asking a system to judge its own performance. In traditional software engineering, we rely on deterministic verification: a function either returns the expected output or it does not. The logic is binary, the test cases are finite, and the compiler is an impartial referee. But [...]

There's a quiet moment in every machine learning engineer's career when they first encounter Reinforcement Learning from Human Feedback (RLHF). It feels like a revelation. The standard supervised learning paradigm, where we meticulously label static datasets, suddenly seems primitive in comparison. RLHF appears to be the missing link, the mechanism that bridges the gap between [...]

The Brittle Promise of Perfect Alignment There is a specific kind of quiet that settles in when a large language model produces something truly uncanny. It isn’t the obvious "AI-isms" of a few years ago—the over-formal tone or the bizarrely repetitive phrasing. It is something subtler: a response that is technically correct, perfectly formatted, and [...]

Building an evaluation pipeline for AI systems often feels like a paradox. We are trying to automate the assessment of intelligence, a concept that resists rigid definition, using code that is inherently deterministic. When I first started deploying machine learning models into production environments, I treated evaluation as a final checkbox before deployment—run a few [...]

The allure of building a state-of-the-art AI model is undeniable, but the real engineering magic—and the source of genuine trust in these systems—lies in how we measure them. We often obsess over architectural tweaks and hyperparameter tuning, yet the evaluation pipeline is frequently an afterthought, cobbled together with a few standard metrics and a validation [...]

When we talk about AI hallucinations, the conversation often defaults to the user's responsibility: "be more specific in your prompt," "use few-shot examples," "provide better context." While these are valid strategies, they place the entire burden of reliability on the person interacting with the model, not the one building it. For engineers deploying Large Language [...]

When we discuss the fragility of Large Language Models (LLMs), the term "hallucination" often feels misleadingly poetic. It suggests a model possessing a mind that can wander or dream. In reality, what we observe is a deterministic mathematical failure: a statistical model assigning high probability to sequences of tokens that do not align with grounded [...]

Artificial intelligence systems are no longer just tools; they are becoming collaborators, decision-makers, and autonomous agents embedded in the critical infrastructure of our digital lives. As these systems grow in capability, they also grow in complexity, opacity, and potential for failure. Traditional software development relies on rigorous testing, but testing for AI is fundamentally different. [...]

There's a pervasive myth in the startup world, particularly among engineering teams moving at light speed, that security is a perimeter problem. We build our walls high, install sophisticated gates in the form of firewalls and authentication layers, and assume that whatever happens inside the fortress is inherently safe. When it comes to traditional software, [...]

When boardrooms discuss artificial intelligence, the conversation often orbits around efficiency gains, competitive advantage, and the sheer novelty of the technology. While these are valid points, they represent only the visible surface of a massive, submerged structure. Beneath the glossy promise of automation lies a complex web of risks that can fundamentally destabilize an organization. [...]

The Illusion of the "Safe" Deployment There is a pervasive, almost seductive narrative currently making the rounds in boardrooms across the globe. It suggests that Artificial Intelligence, particularly Generative AI, is simply another productivity tool—a faster typewriter, a smarter calculator, a digital intern that requires little more than a subscription fee and a basic acceptable [...]

The question of who owns correctness in an artificial intelligence system is deceptively simple. In traditional software engineering, we have established paradigms for accountability. A backend engineer owns the API contract; a database administrator owns the schema integrity; a frontend developer owns the rendering logic. The lines are drawn, the unit tests are written, and [...]

There’s a peculiar tension that surfaces in almost every AI team I’ve worked with or observed. It usually starts with a seemingly innocuous question: "Is this model working correctly?" What follows is rarely a simple technical check. Instead, it triggers a cascade of ownership disputes that span code, data, business logic, and ultimately, the definition [...]

There's a pervasive myth in the technology sector, a ghost that haunts boardrooms and hiring committees alike: the idea that a sufficiently talented data scientist can conjure a production-ready AI product from raw data and computational power alone. We see the job postings demanding "Python wizardry," "Mastery of PyTorch," and "Expertise in NLP," as if [...]

If you spend enough time around AI product teams, you’ll inevitably hear a certain kind of frustration. It usually starts with a data scientist showing off a model with breathtaking accuracy on a validation set, only for the product manager to ask a simple question: "So, can we ship it next Tuesday?" The silence that [...]

The history of artificial intelligence is often told as a story of algorithms, neural networks, and raw computational power. We celebrate the architects of large language models and the researchers pushing the boundaries of reinforcement learning. Yet, beneath the surface of these headline-grabbing advancements lies a quieter, more foundational discipline that has been the bedrock [...]

When we talk about artificial intelligence, the conversation almost immediately drifts toward the towering achievements of Large Language Models, the uncanny realism of generative image systems, or the race toward Artificial General Intelligence (AGI). We marvel at the sheer scale of parameters and the terabytes of data digested during training. Yet, beneath the surface of [...]