AI Resume + Portfolio Optimizer

Project context

Build a portfolio-grade assistant that scores resume quality, pinpoints weak bullets, and produces role-aligned rewrite outputs with measurable claims. Recruiters and applicant tracking systems (ATS) both reward the same underlying signal: specific, quantified impact tied to the target role, not generic self-description. This project teaches you to turn that insight into a repeatable, auditable pipeline instead of a one-off manual edit. Most people edit their resume by feel: they reread it, cringe at a vague line, and rewrite it based on instinct. That approach does not scale, is not measurable, and produces different results every time. This project replaces instinct with a system. You will build a small but real "resume quality engine" made of three connected parts: a scoring function that turns fuzzy quality into numbers, a rewriting process that targets the weakest numbers first, and an audience-aware summarizer that adapts the same underlying facts for different readers. This mirrors how production content-quality tools actually work, whether they are grading essays, scoring ad copy, or evaluating support tickets: define the rubric, score against it, generate targeted fixes, and re-score to prove the fix worked. Along the way you will practice several transferable skills that show up far beyond resumes. Rubric design teaches you how to convert a subjective judgment ("this resume feels weak") into an explicit, defensible set of criteria that two different reviewers would score the same way. Deterministic scoring functions teach you to resist the temptation to hide logic inside a single opaque prompt, keeping your reasoning inspectable and debuggable in plain code. Bullet rewriting teaches the action-scope-metric-outcome pattern, which is the written-communication equivalent of the STAR interview method and shows up constantly in performance reviews, promotion packets, and case studies -- not just resumes. Audience-specific summarization teaches you that the same underlying facts often need different framing for different readers, a skill central to technical writing, product marketing, and even leadership communication. You will also confront a few realistic failure modes on purpose. What happens when a metric is not verified -- do you invent one, or flag it as a placeholder? What happens when two rubric dimensions conflict, such as conciseness versus completeness? How do you know your "improvement" is real and not just a scoring-method artifact from comparing two different rubric versions? Working through these tensions is what separates a toy exercise from a project you can actually defend in an interview. By the end, you will have a tool you can rerun on any future resume draft, a clear before/after score delta with a plain-language explanation of what changed and why, and a portable mental model -- score, target the weak spot, rewrite, re-score -- that you will reuse in almost every later project in this track, including the churn model, the conversion analyzer, and the finance coach.