[Bug]: Duplicate skills inflate recommendation score — parse_skills() doesn't deduplicate

[sujitsingh8]

What happened?

parse_skills() in recommender.py doesn't remove duplicate skills. If a user enters the same skill multiple times (like "python, python, python"), each copy is counted as a separate match in score_single_project, multiplying the skill score. This lets a project's score be inflated just by repeating a skill, which distorts the ranking. Also "py" and "python" both normalize to "python" but are still counted as two separate skills.

Steps to reproduce

1. Call get_recommendations("python", "Beginner", "Web", "Low") and note the scores
2. Call get_recommendations("python,python,python", "Beginner", "Web", "Low")
3. The repeated skill multiplies the score (a single match gives 8, three copies give 14)
4. parse_skills("python, python, py") returns ["python","python","python"] instead of ["python"]

Expected behaviour

Duplicate skills should be counted only once. parse_skills() should return a deduplicated list so repeating a skill can't inflate the recommendation score or change the ranking.

Area of the app affected

Recommendation results

Python version

3.14

Operating system

Windows 11

Relevant error output or logs

No error thrown — silent ranking bug. parse_skills("python, python, py") returns ['python', 'python', 'python'].

Before submitting

• I searched existing issues and this has not been reported before.
• I can reproduce this bug consistently with the steps above.
• I am running the latest version of the main branch.

[varsha-lab2028]

Hi @komalharshita! I'd love to take this one.

I've looked at the bug — parse_skills() splits and normalizes the input but doesn't deduplicate, so repeated entries like "python, python, python" all pass through and inflate the score in score_single_project().

The fix is straightforward: after normalizing each skill, convert to a set before returning, so both exact duplicates and aliases that normalize to the same string are deduplicated in one step.

I'll also add tests covering duplicate input and alias normalization to make sure the fix holds.

Could you please assign issue #710 to me? Happy to open a PR shortly after.