A five-student Loughborough University team won the Microsoft Embrace x Midlands Hackathon 2026 at Loughborough’s campus on May 11 with Lighthouse, an Azure AI-powered careers platform built in five hours to help students map strengths, skills gaps, and future pathways. The victory is a small campus story with a bigger Windows-era theme: Microsoft’s AI push is no longer confined to Copilot demos, enterprise licensing, or cloud strategy decks. It is being handed to students as a practical design constraint, a prompt to solve familiar problems with production-grade tools. In this case, the problem was not “how do we use AI?” but “why do so many students wait until panic sets in before planning a career?”
The Microsoft Embrace x Midlands Hackathon 2026 did not produce a new operating system, a new device category, or a headline-grabbing enterprise product. It produced something more revealing: a student-built prototype aimed at the awkward, anxious, often deferred business of career planning.
Lighthouse, the winning platform, uses Microsoft Azure AI to conduct an interactive interview with users, identify their strengths and interests, surface skills gaps, compare possible career paths, and recommend modules, courses, work experience, and development opportunities. That is a neatly contained use case, but it is also exactly the kind of use case Microsoft wants Azure AI to inhabit: not magic, not novelty, but a workflow that feels underserved by existing software.
The project came from Heechan Yang, a second-year Computer Science and AI student, working with Charlot Eberlein from Computer Science, Antrea Antonia Mavrommati from Product Design Engineering, Hanaa Babrakar from Commercial Management and Quantity Surveying, and Lena Krämer from Robotics. That mix matters. Lighthouse is not merely a computer science exercise; it is a product design, education, and decision-support exercise wrapped in a cloud AI implementation.
The most interesting detail is not that the team used AI. It is that they reportedly resisted starting with AI. According to the university’s account, the team first discussed common student problems and only then settled on career planning as the issue to tackle.
That sequence is the difference between a plausible product and a hackathon cliché. In 2026, “AI-powered” is easy to attach to anything. The harder discipline is finding a task where a conversational system, recommendation engine, and structured guidance might actually reduce friction rather than add another dashboard to the pile.
A careers service can run workshops, publish guides, offer appointments, and maintain employer links. Those resources can be excellent and still miss the moment when a first- or second-year student quietly wonders whether their degree is pointing toward the future they imagined. The issue is not always lack of information. Often it is lack of timing, confidence, and translation.
That is where Lighthouse’s design makes intuitive sense. An interactive interview can meet a student at an earlier, lower-stakes moment. It can ask what they enjoy, what they avoid, what they have tried, and what they do not yet know how to describe. Then it can turn that messy self-assessment into something more actionable.
The product’s recommendation layer is the crucial piece. If a student says they are interested in robotics, sustainability, product design, or commercial project management, a good system should not simply output job titles. It should connect those interests to modules, courses, work placements, portfolio work, societies, internships, and near-term choices.
That is a much more useful form of AI than a chatbot that says, “You might enjoy being a consultant.” Students do not need career fortune-telling. They need a ladder of next steps.
Still, five hours is enough to show what has changed in software development. A small, multidisciplinary student team can now stitch together cloud AI capabilities, a front-end experience, and a coherent product story in an afternoon. That would have been much harder a decade ago, and it would have required more custom machine-learning work even a few years ago.
The platform’s name, Lighthouse, is apt in the way good hackathon names often are. It suggests guidance without pretending to be the destination. It frames the system as a navigational aid rather than an oracle.
That distinction will matter if the team develops it further before presenting at Microsoft’s offices later this summer. The best version of Lighthouse will not tell students who they are. It will help them inspect what they know about themselves, compare options, and take practical steps earlier than they otherwise would have.
The worst version would become another black box dressed up as personalisation. If the system recommends a path, students will need to understand why. If it identifies a skills gap, it should distinguish between “you have not studied this yet,” “you have not demonstrated this yet,” and “the system has insufficient evidence.”
This is where Azure AI’s enterprise sheen cuts both ways. Microsoft’s cloud stack gives developers access to powerful language and recommendation capabilities, but the presence of a major vendor does not automatically solve the human problem. A careers tool must be explainable enough to earn trust and humble enough to admit uncertainty.
Bias is the obvious concern, but not the only one. If a system learns from historical career pathways, it may reproduce old assumptions about who belongs in which field. If it uses incomplete self-reported data, it may mistake modesty for lack of ability. If it overvalues easily measured skills, it may underplay curiosity, resilience, teamwork, and taste.
There is also the question of institutional responsibility. If a university-backed tool recommends modules or work experience, is it advice, guidance, or merely information? If students act on it, who is accountable for the quality of the recommendation?
None of this invalidates Lighthouse. It makes the project more interesting. A serious AI careers platform would have to be designed not only for usefulness, but for restraint.
That is exactly where AI can be useful when applied carefully. Students often do not know what to ask. They may not know which skills are transferable, which careers map to their degree, which modules help them move in a new direction, or how to sequence experience over several years.
A conversational interface can lower the barrier to starting. A recommendation system can turn broad ambitions into next actions. A comparison tool can help students see trade-offs between pathways rather than treating career choice as a single irreversible declaration.
But the value is not in replacing careers advisers. It is in making better use of the time before and between human conversations. A student who has already reflected on interests, mapped gaps, and considered several pathways may arrive at a careers appointment with sharper questions.
That is the practical AI pattern universities should watch. The win is not “AI instead of support.” The win is AI as pre-work, triage, reflection, and navigation.
The Microsoft Embrace x Midlands Hackathon was organised by Microsoft in partnership with the Digital Technologies Network and brought together students from universities across the Midlands. The event challenged participants to build AI-powered solutions within a single day. That format compresses product discovery, technical implementation, and pitching into a sprint.
For Microsoft, these events are not charity alone. They are ecosystem work. A student who builds with Azure AI in a hackathon learns not just the abstract idea of cloud AI, but the workflow, defaults, vocabulary, and limitations of Microsoft’s platform.
That kind of exposure compounds. Today’s hackathon prototype can become tomorrow’s portfolio project, dissertation, startup experiment, or enterprise proof of concept. Even when the prototype disappears, the tool familiarity remains.
For WindowsForum readers, this is worth noticing because Microsoft’s AI strategy is not just coming through Windows Update or Copilot buttons. It is moving through universities, developer communities, partner networks, and student competitions. The AI platform war is being fought in classrooms as well as boardrooms.
The old stereotype of a hackathon is a cluster of coders building a tool for other coders. Lighthouse points in a different direction. The problem domain is educational and personal; the solution depends on interface design, user research, recommendation logic, and a credible understanding of student behaviour.
That matters because many AI failures are not model failures. They are framing failures. The system answers the wrong question, optimises for the wrong metric, or assumes users behave more rationally than they do.
A student careers tool cannot simply ask, “What job do you want?” Many students do not know. It needs to ask better upstream questions: what kind of work gives you energy, what skills do you want to build, which constraints matter, what subjects have surprised you, and what experiences have changed your mind?
Those are product questions before they are AI questions. The team’s reported decision to focus on a genuine peer problem before generating concepts is therefore not a minor anecdote. It is the reason the project sounds more credible than the average AI pitch.
But personalisation can also become paternalism if the system’s recommendations harden into identity labels. A student who receives a confident AI-generated pathway may treat it as validation or discouragement, depending on what it says. That gives designers a responsibility to keep options open.
Lighthouse will need to frame its outputs as possibilities, not verdicts. It should help students compare routes rather than rank their futures from best to worst. It should surface adjacent paths and unconventional combinations, not merely funnel users toward the most obvious career title attached to their degree.
A good implementation would also let users challenge the system. If a student dislikes a recommendation, the tool should learn why. If a student wants to explore a stretch goal, the system should show the development path rather than quietly steering them back toward what looks statistically likely.
This is especially important in widening participation contexts. AI guidance must not become a polite mechanism for lowering ambition. The best careers platform is one that makes ambition more navigable, not one that makes aspiration more predictable.
But university software is not exempt from the hard questions that enterprise IT asks every day. What data is collected during the interview? How long is it retained? Is it used to improve the system? Can students delete it? Are recommendations auditable? How are model outputs evaluated for accuracy and fairness?
Those questions may sound premature for a hackathon prototype, but they become unavoidable if the project moves toward real deployment. Careers data can be sensitive because it touches academic performance, confidence, socioeconomic constraints, immigration status, disability accommodations, and personal aspirations.
There is also the matter of integration. To recommend relevant modules or opportunities well, Lighthouse would need accurate, current institutional data. That means connecting to course catalogues, careers databases, placement systems, event listings, and perhaps student records.
The technical challenge, then, is not merely generating fluent advice. It is maintaining reliable context. In AI systems, stale or incomplete data can produce confident nonsense, and career planning is not a place where confident nonsense is harmless.
That sentiment could easily become motivational wallpaper, but in this context it matters. Student innovation is often narrated as sudden brilliance. Someone has an idea, builds a prototype, wins a prize, and the story ends with smiling photos.
The reality is more iterative. People learn how to scope problems, divide work, pitch clearly, recover from awkward demos, and distinguish a clever feature from a useful product. The fourth hackathon win is built partly from the first three disappointments.
That is also a useful corrective to the way AI tools are marketed. Generative AI can make software creation feel instant, but judgment still takes practice. Knowing what not to build is learned through failure.
The Lighthouse team’s win, then, is not just about a platform. It is about a group of students becoming better at turning a vague human problem into a product-shaped argument under time pressure.
Universities are under pressure to show that degrees lead somewhere. Students are under pressure to justify tuition, debt, and time. Employers complain about skills gaps while students struggle to translate coursework into workplace narratives.
A tool like Lighthouse sits at the intersection of those pressures. It could help students make earlier decisions, help careers services reach more people, and help universities demonstrate structured support for employability.
But that same positioning makes the product politically delicate. If a university recommends a platform that nudges students toward certain modules or career pathways, it may influence demand across departments. If employers are integrated into the opportunity pipeline, questions of fairness and access follow.
The challenge is to build a system that supports exploration without quietly becoming a sorting machine. That is harder than building a chatbot, and far more valuable.
Microsoft’s AI Pitch Lands Better When It Leaves the Keynote Stage
The Microsoft Embrace x Midlands Hackathon 2026 did not produce a new operating system, a new device category, or a headline-grabbing enterprise product. It produced something more revealing: a student-built prototype aimed at the awkward, anxious, often deferred business of career planning.Lighthouse, the winning platform, uses Microsoft Azure AI to conduct an interactive interview with users, identify their strengths and interests, surface skills gaps, compare possible career paths, and recommend modules, courses, work experience, and development opportunities. That is a neatly contained use case, but it is also exactly the kind of use case Microsoft wants Azure AI to inhabit: not magic, not novelty, but a workflow that feels underserved by existing software.
The project came from Heechan Yang, a second-year Computer Science and AI student, working with Charlot Eberlein from Computer Science, Antrea Antonia Mavrommati from Product Design Engineering, Hanaa Babrakar from Commercial Management and Quantity Surveying, and Lena Krämer from Robotics. That mix matters. Lighthouse is not merely a computer science exercise; it is a product design, education, and decision-support exercise wrapped in a cloud AI implementation.
The most interesting detail is not that the team used AI. It is that they reportedly resisted starting with AI. According to the university’s account, the team first discussed common student problems and only then settled on career planning as the issue to tackle.
That sequence is the difference between a plausible product and a hackathon cliché. In 2026, “AI-powered” is easy to attach to anything. The harder discipline is finding a task where a conversational system, recommendation engine, and structured guidance might actually reduce friction rather than add another dashboard to the pile.
Lighthouse Understands the Timing Problem
Heechan Yang’s explanation of the idea cuts to the product’s real thesis: too many students leave career planning until their final year, when pressure suddenly builds. That is not a software bug in university life, but it is a systems problem. Students are asked to make long-horizon decisions while living inside short-horizon academic calendars.A careers service can run workshops, publish guides, offer appointments, and maintain employer links. Those resources can be excellent and still miss the moment when a first- or second-year student quietly wonders whether their degree is pointing toward the future they imagined. The issue is not always lack of information. Often it is lack of timing, confidence, and translation.
That is where Lighthouse’s design makes intuitive sense. An interactive interview can meet a student at an earlier, lower-stakes moment. It can ask what they enjoy, what they avoid, what they have tried, and what they do not yet know how to describe. Then it can turn that messy self-assessment into something more actionable.
The product’s recommendation layer is the crucial piece. If a student says they are interested in robotics, sustainability, product design, or commercial project management, a good system should not simply output job titles. It should connect those interests to modules, courses, work placements, portfolio work, societies, internships, and near-term choices.
That is a much more useful form of AI than a chatbot that says, “You might enjoy being a consultant.” Students do not need career fortune-telling. They need a ladder of next steps.
The Five-Hour Build Shows Both Promise and the Limits of the Demo
Lighthouse was built in just five hours, which is both impressive and a warning label. Hackathon prototypes are designed to persuade judges quickly. They are not designed to survive procurement reviews, accessibility testing, data protection assessments, model evaluation, abuse cases, or the beautifully destructive habits of real users.Still, five hours is enough to show what has changed in software development. A small, multidisciplinary student team can now stitch together cloud AI capabilities, a front-end experience, and a coherent product story in an afternoon. That would have been much harder a decade ago, and it would have required more custom machine-learning work even a few years ago.
The platform’s name, Lighthouse, is apt in the way good hackathon names often are. It suggests guidance without pretending to be the destination. It frames the system as a navigational aid rather than an oracle.
That distinction will matter if the team develops it further before presenting at Microsoft’s offices later this summer. The best version of Lighthouse will not tell students who they are. It will help them inspect what they know about themselves, compare options, and take practical steps earlier than they otherwise would have.
The worst version would become another black box dressed up as personalisation. If the system recommends a path, students will need to understand why. If it identifies a skills gap, it should distinguish between “you have not studied this yet,” “you have not demonstrated this yet,” and “the system has insufficient evidence.”
Career Advice Is a High-Stakes Place to Put a Black Box
There is a reason AI careers platforms deserve scrutiny, even when they arrive as well-intentioned student projects. Career guidance affects confidence, money, time, identity, and opportunity. A bad recommendation can be merely annoying in a music app; in education, it can quietly narrow a student’s imagination.This is where Azure AI’s enterprise sheen cuts both ways. Microsoft’s cloud stack gives developers access to powerful language and recommendation capabilities, but the presence of a major vendor does not automatically solve the human problem. A careers tool must be explainable enough to earn trust and humble enough to admit uncertainty.
Bias is the obvious concern, but not the only one. If a system learns from historical career pathways, it may reproduce old assumptions about who belongs in which field. If it uses incomplete self-reported data, it may mistake modesty for lack of ability. If it overvalues easily measured skills, it may underplay curiosity, resilience, teamwork, and taste.
There is also the question of institutional responsibility. If a university-backed tool recommends modules or work experience, is it advice, guidance, or merely information? If students act on it, who is accountable for the quality of the recommendation?
None of this invalidates Lighthouse. It makes the project more interesting. A serious AI careers platform would have to be designed not only for usefulness, but for restraint.
The Best AI Products Start With a Boring Human Problem
The strongest part of the Lighthouse story is that the team appears to have found a boring problem before building an exciting demo. Career uncertainty is not a glamorous hackathon theme. It does not have the cinematic urgency of disaster response or the futuristic sheen of autonomous agents. But it is widespread, emotionally real, and operationally messy.That is exactly where AI can be useful when applied carefully. Students often do not know what to ask. They may not know which skills are transferable, which careers map to their degree, which modules help them move in a new direction, or how to sequence experience over several years.
A conversational interface can lower the barrier to starting. A recommendation system can turn broad ambitions into next actions. A comparison tool can help students see trade-offs between pathways rather than treating career choice as a single irreversible declaration.
But the value is not in replacing careers advisers. It is in making better use of the time before and between human conversations. A student who has already reflected on interests, mapped gaps, and considered several pathways may arrive at a careers appointment with sharper questions.
That is the practical AI pattern universities should watch. The win is not “AI instead of support.” The win is AI as pre-work, triage, reflection, and navigation.
Hackathons Are Becoming Microsoft’s Grassroots AI Channel
Microsoft has spent the past several years positioning AI as a layer across Windows, Microsoft 365, GitHub, Azure, security tooling, and developer platforms. But hackathons like this one show another route into the market: put the tools in front of students before they become developers, founders, IT managers, or procurement decision-makers.The Microsoft Embrace x Midlands Hackathon was organised by Microsoft in partnership with the Digital Technologies Network and brought together students from universities across the Midlands. The event challenged participants to build AI-powered solutions within a single day. That format compresses product discovery, technical implementation, and pitching into a sprint.
For Microsoft, these events are not charity alone. They are ecosystem work. A student who builds with Azure AI in a hackathon learns not just the abstract idea of cloud AI, but the workflow, defaults, vocabulary, and limitations of Microsoft’s platform.
That kind of exposure compounds. Today’s hackathon prototype can become tomorrow’s portfolio project, dissertation, startup experiment, or enterprise proof of concept. Even when the prototype disappears, the tool familiarity remains.
For WindowsForum readers, this is worth noticing because Microsoft’s AI strategy is not just coming through Windows Update or Copilot buttons. It is moving through universities, developer communities, partner networks, and student competitions. The AI platform war is being fought in classrooms as well as boardrooms.
The Multidisciplinary Team Is the Real Product Signal
The Lighthouse team’s composition says something important about where applied AI is heading. Computer science and AI expertise were present, but so were product design engineering, commercial management and quantity surveying, and robotics. That is closer to how useful AI products are actually built.The old stereotype of a hackathon is a cluster of coders building a tool for other coders. Lighthouse points in a different direction. The problem domain is educational and personal; the solution depends on interface design, user research, recommendation logic, and a credible understanding of student behaviour.
That matters because many AI failures are not model failures. They are framing failures. The system answers the wrong question, optimises for the wrong metric, or assumes users behave more rationally than they do.
A student careers tool cannot simply ask, “What job do you want?” Many students do not know. It needs to ask better upstream questions: what kind of work gives you energy, what skills do you want to build, which constraints matter, what subjects have surprised you, and what experiences have changed your mind?
Those are product questions before they are AI questions. The team’s reported decision to focus on a genuine peer problem before generating concepts is therefore not a minor anecdote. It is the reason the project sounds more credible than the average AI pitch.
Personalisation Is Powerful Only When It Does Not Become Paternalism
The word “personalised” now appears in nearly every AI product pitch, and for good reason. Generic guidance is easy to ignore. A system that reflects a user’s interests, strengths, and gaps feels more relevant and more motivating.But personalisation can also become paternalism if the system’s recommendations harden into identity labels. A student who receives a confident AI-generated pathway may treat it as validation or discouragement, depending on what it says. That gives designers a responsibility to keep options open.
Lighthouse will need to frame its outputs as possibilities, not verdicts. It should help students compare routes rather than rank their futures from best to worst. It should surface adjacent paths and unconventional combinations, not merely funnel users toward the most obvious career title attached to their degree.
A good implementation would also let users challenge the system. If a student dislikes a recommendation, the tool should learn why. If a student wants to explore a stretch goal, the system should show the development path rather than quietly steering them back toward what looks statistically likely.
This is especially important in widening participation contexts. AI guidance must not become a polite mechanism for lowering ambition. The best careers platform is one that makes ambition more navigable, not one that makes aspiration more predictable.
The Azure Angle Brings Enterprise Questions Into Student Software
Because Lighthouse uses Microsoft Azure AI, it sits within a broader enterprise conversation about trust, governance, and deployment. The underlying platform choice offers practical advantages: scalable AI services, integration potential, and a vendor ecosystem already familiar to many institutions.But university software is not exempt from the hard questions that enterprise IT asks every day. What data is collected during the interview? How long is it retained? Is it used to improve the system? Can students delete it? Are recommendations auditable? How are model outputs evaluated for accuracy and fairness?
Those questions may sound premature for a hackathon prototype, but they become unavoidable if the project moves toward real deployment. Careers data can be sensitive because it touches academic performance, confidence, socioeconomic constraints, immigration status, disability accommodations, and personal aspirations.
There is also the matter of integration. To recommend relevant modules or opportunities well, Lighthouse would need accurate, current institutional data. That means connecting to course catalogues, careers databases, placement systems, event listings, and perhaps student records.
The technical challenge, then, is not merely generating fluent advice. It is maintaining reliable context. In AI systems, stale or incomplete data can produce confident nonsense, and career planning is not a place where confident nonsense is harmless.
The Win Is Also a Story About Not Stopping
Hackathons tend to celebrate speed, but Heechan Yang’s post-win reflection points to persistence. He said this was his fourth hackathon, after three that did not bring the result he wanted. His lesson was blunt: the first three times not working out did not mean he was bad at hackathons; it meant he had not won yet.That sentiment could easily become motivational wallpaper, but in this context it matters. Student innovation is often narrated as sudden brilliance. Someone has an idea, builds a prototype, wins a prize, and the story ends with smiling photos.
The reality is more iterative. People learn how to scope problems, divide work, pitch clearly, recover from awkward demos, and distinguish a clever feature from a useful product. The fourth hackathon win is built partly from the first three disappointments.
That is also a useful corrective to the way AI tools are marketed. Generative AI can make software creation feel instant, but judgment still takes practice. Knowing what not to build is learned through failure.
The Lighthouse team’s win, then, is not just about a platform. It is about a group of students becoming better at turning a vague human problem into a product-shaped argument under time pressure.
The Campus Prototype Points to a Bigger Market
Higher education is crowded with advising tools, learning platforms, student-success dashboards, and employability services. Lighthouse would not enter an empty field. But AI changes the competitive shape of that field by making interactive guidance easier to prototype and potentially easier to scale.Universities are under pressure to show that degrees lead somewhere. Students are under pressure to justify tuition, debt, and time. Employers complain about skills gaps while students struggle to translate coursework into workplace narratives.
A tool like Lighthouse sits at the intersection of those pressures. It could help students make earlier decisions, help careers services reach more people, and help universities demonstrate structured support for employability.
But that same positioning makes the product politically delicate. If a university recommends a platform that nudges students toward certain modules or career pathways, it may influence demand across departments. If employers are integrated into the opportunity pipeline, questions of fairness and access follow.
The challenge is to build a system that supports exploration without quietly becoming a sorting machine. That is harder than building a chatbot, and far more valuable.
The Lighthouse Lesson for Microsoft’s AI Era
The concrete lesson from Loughborough is not that every student service needs a chatbot. It is that the useful frontier for AI is often found in the spaces where people already feel stuck, embarrassed, late, or overloaded. Career planning fits that pattern almost perfectly.- Lighthouse won first place at the Microsoft Embrace x Midlands Hackathon 2026 after being built during a single-day event at Loughborough’s campus on May 11.
- The platform uses Microsoft Azure AI to interview students, identify strengths and skills gaps, compare career pathways, and recommend development opportunities.
- The team included students from Computer Science, Computer Science and AI, Product Design Engineering, Commercial Management and Quantity Surveying, and Robotics.
- The project’s strongest product decision was to begin with a real student problem rather than starting with AI for its own sake.
- Any future version of Lighthouse will need strong safeguards around explainability, data privacy, bias, and the difference between guidance and automated judgment.
- The team’s invitation to present at Microsoft’s offices later this summer turns a campus hackathon win into a live test of whether a five-hour prototype can mature into a credible student-support product.
References
- Primary source: Loughborough University
Published: 2026-06-09T08:30:07.670200
Loughborough students win Microsoft Embrace x Midlands Hackathon with AI-powered careers platform
A team of Loughborough students recently won first place at the Microsoft Embrace x Midlands Hackathon 2026 by developing an AI-powered platform designed to help students make informed decisions about their future careers.
www.lboro.ac.uk