BuildYourPassword

Problem

Password strength UX usually stops at simple rules. Recruiters and users benefit from a system that combines heuristics, breach signals, and trained classifiers.

Approach

• ▸Rule-based scoring for length, complexity, and common patterns
• ▸Optional breach / exposure checks via external API
• ▸Train and compare four classic ML classifiers on labeled features
• ▸Export models with joblib for reproducible inference demos

Decisions

• ▸Flask for a thin API + UI suitable for ML model serving demos
• ▸Keep multiple models side-by-side to discuss tradeoffs in interviews
• ▸Separate notebooks for evaluation from the serving path

Challenges

• ▸Balancing latency vs model ensemble size for interactive use
• ▸Hardened secrets handling before any public deploy
• ▸Serverful Python (gunicorn + sklearn) needs a host beyond Vercel

Results

• ▸Clear comparison story across four model families
• ▸End-to-end pipeline from features → train → serve
• ▸Strong talking points for ML + backend interviews