In February 2025 my team and I took part in Udhgam, a 24-hour hackathon at Woxsen University. It was my first hackathon. Over those 24 hours we went from a blank whiteboard to a working product and a pitch, and we finished as 1st Runner-Up out of more than 130 teams.
The problem
Students and early-career people usually get stuck in one of two places. Some know what they can do but not which roles those skills lead to. Others know the role they want but have no clear path to it. Career advice tends to be generic, and job-market information is scattered across many sites.
We wanted one place that connects skills and career goals in both directions, in a structured and accessible way.
What we built
NexTrail is a career guidance platform built around two journeys:
- Skills to roles: enter the skills you already have, and NexTrail suggests the job roles they fit best.
- Role to roadmap: enter the role you want, and NexTrail generates a step-by-step roadmap of the skills to learn to get there.
- Job trends: an overview of what the industry is currently demanding, so decisions are based on real demand rather than guesswork.
- Community: users can connect with others on similar paths.
How it works
The core is a machine learning model rather than a hard-coded lookup table, so it can generalise to skill combinations it hasn't seen written down anywhere.
- A user's skills are converted into embeddings, numeric vectors that capture what each skill means, so related skills end up close together.
- A custom neural network, built and trained in PyTorch, takes those skill embeddings and maps them to the most relevant job roles.
- A Flask backend serves the model through REST inference APIs that the frontend calls.
- For the role-to-roadmap journey, an LLM (Mistral Small 24B, from Hugging Face) generates a personalised learning plan, which is rendered as a flow visualization: an ordered path of skills rather than a wall of text.
The 24 hours
Most of the challenge was scope. Twenty-four hours goes quickly, so we had to decide early what the demo absolutely needed: the skill-to-role model working end to end, the roadmap generation, and a clean flow to present. Everything else came after that.
The day ran from brainstorming and ideation, through building the model, the APIs and the interface in parallel, to integrating everything and finally pitching it to the judges. Keeping the pieces in sync while each person worked on a different part was as much of a challenge as the code itself.
The team
Each of us brought different strengths, and the collaboration is what turned the idea into something that worked on stage:
What I took away
As a first hackathon, it taught me things that are hard to learn alone: how to cut a big idea down to a demo-able core, how to split work across a team without stepping on each other, and how to explain a technical system to judges in a few minutes.
Thanks to Woxsen University for organising Udhgam, and to the mentors and fellow participants who made it such a good 24 hours. It left me wanting to do many more.