Project case study
Resume Studio — AI Resume Grader
Grades and upgrades a visitor's resume without turning the portfolio into a generic utility dashboard.
Problem
A public resume workstation for ATS grading, optional job matching, grounded AI rewriting, and LaTeX plus PDF export.
Context
Built as a standalone live project so visitors can grade and upgrade their own resumes without distracting from the portfolio's core hiring narrative.
My Role
AI/ML Engineer, Data Systems Builder & Motion UI Developer
Contribution
- Implemented role-description analysis with keyword matching, ATS checks, and a transparent fit score.
- Added optional role matching while keeping a useful general score when no job description is supplied.
- Integrated NVIDIA NIM behind explicit consent and server-only credentials, with measurable-claim grounding and deterministic fallback.
- Built escaped, compile-ready LaTeX and an independent styled PDF export path.
Architecture
Role and resume input
Collects a role description and either canonical or visitor resume content.
Feeds Evidence analyser
Evidence analyser
Calculates keyword fit and selects only canonical projects and proof.
Feeds Package generator
Package generator
Prepares tailored copy, LaTeX, PDF, and optional outreach content.
Feeds Review and delivery
Review and delivery
Keeps exports reviewable before download or configured email delivery.
Terminal stage
Engineering Decisions
Ground generated resume content in canonical portfolio records
Reason: Role tailoring should reorder and phrase verified work without inventing experience.
Tradeoff: The output is constrained by the completeness of the source archive.
Keep deterministic analysis alongside optional provider-backed generation
Reason: Keyword and ATS feedback must remain useful when an AI provider is unavailable.
Tradeoff: The deterministic path is less expressive than a reviewed provider-generated draft.
Outcomes
No measured outcome is documented in the current source archive.
Limitations
- Provider-backed generation and email delivery depend on valid server configuration and fail closed when unavailable.
- The fit score is guidance for review, not a promise of recruiter or ATS outcomes.
Stack
- Next.js
- TypeScript
- Prompt Design
- NVIDIA NIM
- LaTeX
Evidence
No external evidence is linked to this project.