How to choose your final-year project domain (without regretting it in March)

Every January we meet teams who picked a domain in September for the wrong reason — a senior's leftover code, a trending buzzword, or simply whatever the first search result suggested. By March they're rewriting abstracts at midnight. Choosing well in the beginning costs one honest week; choosing badly costs the whole semester.

Start with your evaluation panel, not the internet. Read your department's project rubric and last year's approved title list. Panels reward clear problem statements, measurable results, and a working demo far more than they reward exotic keywords. A well-executed IoT irrigation controller with real field data beats a half-trained transformer model every time.

Second circle: feasibility with your actual resources. Count what you truly have — lab access hours, a GPU or not, budget for components, and team members' current skills (not aspirational ones). A deep-learning project without a GPU plan is a waiting-room project. A hardware project without a component budget and a soldering-capable teammate is a simulation project wearing a costume.

Third circle: your placement narrative. Interviewers ask 'walk me through your project' in nearly every campus round. Pick a domain you'd enjoy defending for ten minutes — where you can explain the why, the architecture, the hardest bug, and what you'd do differently. If the target is software roles, a full-stack or ML project you built end to end tells a stronger story than a domain you can't discuss beyond the abstract.

Where the three circles overlap, shortlist two or three concrete titles and pressure-test each with one question: what is the measurable output? 'Accuracy improved from X to Y on this dataset', 'water usage reduced by Z% across a two-week trial', 'response time under 200 ms at 100 concurrent users'. If you can't phrase the result as a number, the reviewers can't either.

Finally, decide the build-vs-guidance split honestly. There's no shame in getting structured guidance — that's how industry works too — but make sure you can rebuild the core module alone. Your viva, and your first job interview, will check exactly that.