ResumeDojo – AI Resume Intelligence & ATS Evaluation Engine

Job seekers face high ATS rejection rates and opaque screening algorithms with generic advice. Most resume checkers rely on naive keyword density counting rather than grounded bullet-level impact evaluation, structure validation, role matching, or actionable rewrites.

Engineered an end-to-end intelligence pipeline: (1) Layout-aware text extraction with Docling & PyMuPDF, (2) Deterministic ResumeProfileExtractor normalizing candidate skills, metrics, and timeline into structured Pydantic models, (3) Calibrated multi-factor ATS Composite Scoring Engine with schema repair (RobustPydanticParser), (4) Bullet-level critic analyzing outcome vs. activity orientation and generating quantified rewrites, (5) Canonical role matching across 12+ pre-configured engineering tracks, and (6) Dynamic LaTeX template recommender. Built with FastAPI and Next.js 14.

• →Deterministic structured extraction pipeline parsing complex multi-column PDF resumes
• →Calibrated ATS scoring engine with grounded, actionable, non-contradictory top fixes
• →Bullet-level critic evaluating outcome vs. activity orientation and generating quantified rewrites
• →Automated role-fit matching across 12+ canonical software and ML engineering tracks
• →Job description gap analyzer comparing candidate experience against custom target roles
• →ATS-friendly LaTeX template recommendations tailored to candidate background

• 01Robust JSON schema repair (json-repair + schema fallback) is essential when parsing structured LLM responses into strict Pydantic models
• 02Dual-parser strategy (Docling for complex layouts, PyMuPDF for fast lightweight text) provides optimal balance between fidelity and speed
• 03Separating resume normalization into a cached profile step prevents redundant token spend across multi-module evaluation passes
• 04Grounded prompt design with explicit negative constraints prevents LLM hallucinations in resume critique