<h3>Who we are</h3>
<p>Helsing is a defence AI company. Our mission is to protect our democracies. We aim to achieve technological leadership, so that open societies can continue to make sovereign decisions and control their ethical standards.</p>
<h3>The role</h3>
<p>At Helsing we deliver AI-based capabilities and the enabling infrastructure that allow semi-autonomous platforms to localise, navigate, and perceive the world in real time. You will have the unique opportunity to shape the future of AI in one of the most challenging sectors, where performance needs to be paired up with high generalisation capabilities and strong robustness against adversarial attacks.
You will be part of a team focused on reinforcement learning, where you will design, train and deploy agents in complex multi-agent environments.</p>
<h3>You should apply if you</h3>
<ul>
<li>Hold MSc in machine learning with a speciality in either reinforcement learning, multi-agent systems, automation and control, or robotics.</li>
<li>Have excellent communication skills and the ability to report and present research findings clearly and efficiently both internally and externally.</li>
<li>Are passionate about keeping up-to-date with current research and enjoy reimplementing / extending papers on state-of-the-art Deep Learning-based approaches.</li>
<li>Possess solid software engineering skills, writing clean and well-structured code in Python and/or languages like Rust, Java, or modern C++, and experience deploying AI software to production including testing, QA, and monitoring.</li>
</ul>
<h3>Nice to have</h3>
<ul>
<li>PhD in reinforcement learning, multi-agent systems, automation and control, robotics, or related fields and have authored publications in top-tier journals and conferences (eg, NeurIPS, ICML, ICLR, AAMAS).</li>
<li>Experience with large-scale RL training.</li>
<li>Experience modelling and training multi-agent controllers using state-of-the-art tools and techniques.</li>
</ul>


AI Research Engineer - Reinforcement Learning

<p>For generations, the editor’s red pen was the ultimate arbiter of quality. It was a tool not of negation, but of elevation. With a few precise marks, it could challenge a weak argument, question a dubious source, and sharpen a brilliant-but-unfocused idea into a masterpiece. The red pen was a symbol of critical judgment, of a human mind applying its full depth of context, nuance, and strategic intent to a piece of work.</p>
<p>Today, we are flooded with work produced not by human hands, but by algorithms. Artificial Intelligence can generate reports, write code, and craft marketing campaigns with breathtaking speed. In our rush to embrace this efficiency, we risk becoming passive consumers of its output, accepting the machine’s first draft as the final word. This presents a critical risk. To truly leverage AI, perhaps we shouldn't abandon the editor’s chair, but instead reclaim it with more vigor than ever. We need a new framework for this new era. Let’s call it <strong>The Red Pen Protocol</strong>.</p>
<h3>From Prompt Engineer to Chief Strategist</h3>
<p>The conversation around AI is often centered on "prompt engineering"—the craft of asking the machine a better question. While this is a powerful and strategic skill in its own right, it primarily optimizes our interaction <em>with</em> the AI. The Red Pen Protocol, by contrast, is a framework that governs our interaction <em>with the AI's output</em>. It shifts our role from that of a mere operator to that of a chief strategist, a final arbiter of quality and purpose.</p>
<p>The protocol is not a checklist for catching errors; it is a series of lenses for applying deep, human-centric judgment. It assumes the AI’s output is a gifted but flawed first draft, a starting point for a much deeper conversation. It is built on a single, powerful assumption: <strong>The AI can provide an answer, but only you can understand the context.</strong></p>
<h3>The Four Lenses of the Red Pen Protocol</h3>
<p>The protocol consists of four distinct lenses through which to audit any significant piece of AI-generated work.</p>
<p><strong>1. The Lens of Assumption: What invisible frames are at play?</strong>
Every AI output is built on a foundation of hidden assumptions—the assumptions in its training data, and the assumptions embedded in your own prompt. The first step of the protocol is to make these assumptions visible.</p>
<ul>
<li><strong>Question it:</strong> What established beliefs does this output reinforce? What alternative viewpoints has it ignored? If I were arguing the exact opposite case, what data would I use?</li>
<li><strong>Action:</strong> Before accepting the AI’s analysis, force it to argue against its own conclusion. Use prompts like, "Now, provide a detailed critique of the analysis above, highlighting its potential blind spots and flawed assumptions."</li>
</ul>
<p><strong>2. The Lens of Intent: Does this align with our ultimate strategic goal?</strong>
An AI can optimize for the task you give it, but it cannot grasp the larger strategic intent behind the task. Its output can be tactically brilliant but strategically disastrous.</p>
<ul>
<li><strong>Question it:</strong> Does this output merely answer the question, or does it advance our core mission? Does it reflect our brand’s voice and values? Could this correct answer lead to a wrong long-term outcome?</li>
<li><strong>Action:</strong> Articulate your strategic intent in one sentence. Read the AI’s output. If there is even a slight dissonance between the two, the work must be revised. This is where human leadership is irreplaceable.</li>
</ul>
<p><strong>3. The Lens of Synthesis: Where are the missing connections?</strong>
AI models are powerful pattern-matchers, but they often struggle with true synthesis—the art of connecting disparate ideas to create something novel. They deliver knowledge, but humans create wisdom. This is a key insight we explored in <a href="/articles/the-polymaths-secret-analogical-thinking">The Polymath's Secret: Cultivating Analogical Thinking in an Age of Specialization</a>.</p>
<ul>
<li><strong>Question it:</strong> What fields or ideas outside of this immediate domain could inform this work? What historical analogy could provide a new perspective? Does this output feel like a simple summary, or does it offer a truly unique insight?</li>
<li><strong>Action:</strong> Feed the AI a concept from a completely unrelated field and ask it to draw a connection. For example, "Apply the principles of ecological succession to this marketing plan." The results may be strange, but they will force a higher level of thinking.</li>
</ul>
<p><strong>4. The Lens of Second-Order Effects: What happens next?</strong>
Finally, the protocol demands that we think beyond the immediate result and anticipate the cascading consequences of deploying the AI’s work. An AI operates in the present; a strategist lives in the future.</p>
<ul>
<li><strong>Question it:</strong> If we implement this, what is the likely response from our competitor? How might our customers misinterpret this? What new problem might this solution create a year from now?</li>
<li><strong>Action:</strong> Use the AI as a simulation engine. Present it with its own output and ask it to "red team" the potential negative consequences. "We are about to launch this marketing copy. Act as our fiercest competitor and describe how you would exploit its weaknesses."</li>
</ul>
<h3>A Note on Proportionality</h3>
<p>It is crucial to apply the Red Pen Protocol with a sense of proportion. It is not a framework for editing a two-sentence email. It is a strategic tool to be deployed when the stakes are high: a major report, a piece of critical code, a new marketing strategy. The goal is not to create friction in your workflow, but to apply focused, high-leverage scrutiny where it matters most. Knowing <em>when</em> to apply the protocol is as important as knowing how.</p>
<h2>Conclusion: The Human in the Loop is the Mind in Charge</h2>
<p>Viewing our work through the Red Pen Protocol is more than a quality control mechanism; it's an exercise in cognitive leadership. It asserts that in the age of AI, the most valuable human skill is not the ability to generate answers, but the wisdom to critically evaluate, strategically align, and thoughtfully elevate them.</p>
<p>The future of professional work is not a battle against the machine, but a partnership with it. But for this partnership to be fruitful, we must be the senior partner. We must be the one who holds the red pen. The AI can be the writer, but we must remain the editor, the strategist, and the final, thoughtful authority.</p>


The Red Pen Protocol: A Strategist’s Guide to Auditing AI

<p>It’s a strange feeling, isn’t it? You see a colleague in a seemingly adjacent role get a promotion that launches them into the stratosphere. You read about a new startup that automates a task you spent years mastering. You feel the ground shifting beneath your feet, a sense that the old rules of career progression—climb the ladder, deepen your niche—are becoming less reliable.</p>
<p>This isn’t chaos. It’s the emergence of a new, predictable force reshaping the professional landscape. A force I call <strong>The Law of Economic Gravity.</strong></p>
<p>For centuries, economic value was anchored to the execution of tasks and the possession of information. If you could do something well or knew something others didn’t, you were valuable. But AI is rapidly commoditizing both execution and information, causing the old center of gravity to collapse. In its place, a new one is forming, pulling value towards a completely different set of human capabilities.</p>
<p>Understanding this shift is the single most important strategic challenge of our time. Navigating it is the key to building a durable, meaningful, and wildly successful career.</p>
<h2>The Three Zones of the New Gravitational Field</h2>
<p>Imagine the new economy as a solar system. Some roles and skills are becoming massive suns, radiating value. Others are becoming cold, forgotten planets. And some are collapsing into the void.</p>
<h3>1. High-Gravity Zones (The Suns)</h3>
<p>These are the domains where economic value is concentrating with immense force. High-Gravity zones are not defined by specific job titles, but by the nature of the work itself. It is work that requires <strong>creative synthesis, complex problem-framing, and high-stakes strategic judgment.</strong></p>
<p>Here, AI is not a replacement; it is a lever. It’s the strategist using AI to model five possible corporate futures, not just to analyze last quarter’s data. It’s the leader using AI to augment team creativity, not just to manage a project plan. It’s the scientist using AI to ask more beautiful, more incisive questions. The gravity here is immense because these are the uniquely human tasks of wisdom, foresight, and architectural thinking that AI cannot replicate.</p>
<h3>2. Low-Gravity Zones (The Planets)</h3>
<p>This is where most professional work currently resides. The work is necessary but increasingly commoditized. It involves predictable, repeatable tasks that can be heavily augmented or partially automated by AI. The work is still necessary but offers little leverage or long-term defensibility.</p>
<p>Think of the analyst who generates a weekly report, the marketer who writes standard ad copy, or the project manager who simply tracks task completion. Residing here isn’t an immediate danger, but it is a long-term risk. There is no "gravity well" to hold you here; you are in a state of perpetual drift, vulnerable to the pull of the void.</p>
<h3>3. Zero-Gravity Zones (The Void)</h3>
<p>This is where skills and roles go when their economic value has been fully extinguished by automation. It’s the silent, empty space once occupied by typesetters, switchboard operators, and, more recently, those performing basic data entry, rote proofreading, or writing simple, boilerplate code.</p>
<p>The pull of the Zero-G void is relentless. The single greatest mistake a professional can make today is to be unaware that their core skills are drifting into this zone.</p>
<h2>The Actionable Framework: Your 3-Step Navigation Plan</h2>
<p>You cannot fight gravity, but you can use it. You can build a vessel that allows you to escape the pull of the void and migrate toward the suns. Here is a simple framework to guide you.</p>
<h3>Step 1: Conduct a Gravitational Audit</h3>
<p>First, map your daily tasks into the three zones. A gravitational audit requires asking honest questions about your work: Does this task require synthesis or just summary? Could 80% of this be done by an AI right now? Is the primary value of this task in execution or in judgment? This audit will give you a clear, and perhaps uncomfortable, picture of where you currently stand.</p>
<h3>Step 2: Initiate a Gravitational Shift</h3>
<p>With a clear map, you can begin to shift your trajectory. Aggressively automate or delegate tasks in the Zero-Gravity zone. For work in the Low-Gravity zone, build escape velocity by focusing on the "why" behind the tasks; move from <em>generating the report</em> to <em>interpreting the report's strategic implications</em>. Finally, actively seek out projects that involve ambiguity, synthesis, and high-stakes decision-making to migrate toward the High-Gravity zones.</p>
<h3>Step 3: Develop Your "Gravity Well"</h3>
<p>A personal "gravity well" is the unique combination of skills and perspectives that creates its own value. This is where cultivating the traits of a "Polymath," practicing "Intellectual Humility," and mastering "Deep Work" become the essential tools for building a high-gravity career. By connecting disparate fields and asking better questions, you create a value proposition that is uniquely your own.</p>
<h2>Conclusion: The Architect of Your Own Economy</h2>
<p>This shift is not about a race against the machine, but about understanding the new physics of the economy and choosing to position yourself where your uniquely human skills shine brightest. It's about moving from being a cog in the old machine to becoming an architect of the new one, designing a career that is not just successful, but durable and meaningful in an age of intelligent machines.</p>


The Law of Economic Gravity: How to Thrive When AI Changes the Value of Everything

<p>We are building machines that can provide us with perfect answers. Given a goal, they can calculate the most efficient path to achieve it with a speed and precision that dwarfs human capability. They can optimize a supply chain, personalize a marketing campaign, or diagnose a disease from a medical scan, all with flawless, data-driven logic. We have, in essence, perfected the science of finding the right path.</p>
<p>But in our obsession with pathfinding, we have neglected the more fundamental human challenge: choosing the right destination. As artificial intelligence automates the world of answers, it reveals a profound and widening chasm—the space between what our technology <em>can</em> do and what it <em>should</em> do. This is the <strong>Ethical Judgment Gap</strong>, and navigating it is the most important work of our time.</p>
<p>The engineer who designs a facial recognition algorithm is a master of optimization. But the decision of whether that algorithm should be used to track dissidents or to find missing children is not a technical problem. It is a moral one. The value of human professionals in the age of AI will be measured not by their ability to solve the problem, but by their wisdom in choosing which problems are worth solving and which solutions are worthy of our humanity.</p>
<h2>The Human Insight: The Navigator's Two Instruments</h2>
<p>Imagine the captain of a 16th-century exploration vessel, standing on the deck of her ship in the vast, featureless expanse of the ocean. She has two critical instruments before her. The first is the astrolabe, a marvel of scientific ingenuity. By measuring the position of the stars, it can tell her, with remarkable precision, her exact latitude. It is a tool of pure data, a perfect calculator of "what is." An AI could master the astrolabe in an instant, providing calculations of infallible accuracy.</p>
<p>But beside it lies her second instrument: a simple, magnetic compass. The compass does not care for the stars. It does not calculate. It does one thing and one thing only: it points north. It provides a constant, unwavering direction. The astrolabe can tell the captain where she is, but only the compass can tell her which way she is going. Her true task as a leader is not merely to read the data from the astrolabe, but to use her judgment to align the ship's course with the unwavering needle of the compass, guided by the ultimate purpose of her mission—be it discovery, trade, or the safe return of her crew.</p>
<p>In the modern world, AI is the ultimate astrolabe. It provides us with a perfect, data-rich picture of our current position. But it cannot provide us with a compass. It cannot tell us which direction is "north." The work of ethical judgment is the work of the captain: to hold the compass, to interpret the data of the astrolabe through the lens of purpose and principle, and to make the courageous choice about which way to steer. This is not a science; it is an art. It is the art of moral navigation.</p>
<h2>A Framework for Moral Navigation</h2>
<p>Wielding a moral compass is not a passive act of virtue; it is an active discipline. Here is a simple, three-step framework for navigating the Ethical Judgment Gap in your own work.</p>
<h3>Step 1: Define Your Compass</h3>
<p>Before you face a complex decision, define your "north." Articulate the core principles that guide you and your organization. These are not project goals; they are foundational values. Write them down. A compass you can't read is useless.</p>
<ul>
<li><strong>Example:</strong> "1. We will not trade user trust for short-term engagement. 2. Our work must empower, not exploit. 3. We will be transparent about our data and algorithms."</li>
</ul>
<h3>Step 2: Consult the Astrolabe</h3>
<p>With your principles clear, engage the full power of the AI. Use it as the ultimate astrolabe to gather data, generate options, and model potential outcomes. Ask it to map the territory with ruthless, data-driven clarity.</p>
<ul>
<li><strong>Example:</strong> "Model the financial impact of removing the addictive feature." or "Generate five alternative solutions that achieve the business goal without collecting user location data."</li>
</ul>
<h3>Step 3: Steer the Course</h3>
<p>This is the art of leadership. Compare the map generated by the astrolabe with the direction of your compass. The optimal path is rarely the most efficient one. The correct choice is the one that best aligns the data-driven options with your predefined principles. This requires judgment, courage, and the willingness to choose a harder right over an easier wrong.</p>
<h2>Conclusion: The Last Defensible Human Skill</h2>
<p>The ability to code, to analyze data, and to manage projects will be increasingly augmented and automated by AI. These are the skills of the astrolabe—the skills of calculating where we are. But the enduring, defensible, and uniquely human skill is the art of wielding the moral compass—the wisdom to decide where we should go.</p>
<p>Developing your capacity for ethical judgment is not a brake on your career or an obstacle to innovation. It is the very engine of sustainable leadership and long-term value creation. It is what will allow you to guide, to build, and to lead in a world that is drowning in answers but starved of wisdom. The AI can help you find the path. Your job is to choose the destination. That is the work that matters. That is the work that will last.</p>


AI Research Engineer - Reinforcement Learning

The Story Behind the Role

Who we are

The role

You should apply if you

Nice to have

Provenance Trail