| title | ResuFit |
|---|---|
| emoji | 🎯 |
| colorFrom | blue |
| colorTo | purple |
| sdk | docker |
| pinned | false |
ResuFit is a production-ready REST API that screens resumes against job descriptions using a hybrid AI scoring model combining TF-IDF semantic similarity and keyword-based skill extraction.
Upload a resume PDF and a job description → receive a structured fit report in milliseconds.
Live API: https://resufit.onrender.com (update after deployment)
Interactive Docs: https://resufit.onrender.com/docs
curl -X POST "https://resufit.onrender.com/api/v1/analyze" \
-F "resume=@alex_johnson_resume.pdf" \
-F "job_description=We are hiring a Python Backend Engineer with FastAPI, Docker, and AWS experience..."Response:
{
"fit_score": 0.6513,
"fit_percentage": 65.1,
"verdict": "Strong Fit",
"advice": "This candidate strongly matches the job requirements. Recommended for interview.",
"matched_skills": ["python", "fastapi", "docker", "aws", "postgresql", "machine learning"],
"missing_skills": ["scikit-learn"],
"extra_skills": ["flask", "redis", "pytorch", "kubernetes"],
"tfidf_score": 0.1964,
"skill_score": 0.9545,
"resume_skill_count": 44,
"jd_skill_count": 22,
"resume_text_length": 3111
}ResuFit uses a two-signal hybrid scoring model designed specifically for technical job matching:
Converts both texts into TF-IDF vectors and computes cosine similarity. Captures:
- Overall semantic alignment between resume and JD
- Seniority language ("architected", "led a team of", "junior developer")
- Domain vocabulary ("distributed systems", "real-time processing")
- Contextual soft skills ("cross-functional", "stakeholder management")
Matches skills against a curated taxonomy of 500+ technical skills across 12 categories using boundary-aware regex patterns. Returns:
- Exact matched skills (present in both resume and JD)
- Missing skills (in JD but absent from resume — the candidate's gaps)
- Extra skills (additional value the candidate brings)
hybrid_score = (tfidf_score × 0.40) + (skill_match_rate × 0.60)
Skills get 60% weight because for technical roles, specific technology matches ("Docker", "FastAPI", "PostgreSQL") are more predictive of job fit than semantic writing similarity.
| Score Range | Verdict |
|---|---|
| ≥ 0.65 | ✅ Strong Fit |
| 0.40 – 0.64 | 🟡 Partial Fit |
| < 0.40 | ❌ Weak Fit |
| Layer | Technology | Purpose |
|---|---|---|
| Web Framework | FastAPI 0.111 | Async REST API, auto-generated docs |
| ML Scoring | Scikit-learn (TF-IDF + Cosine Similarity) | Semantic resume-JD matching |
| PDF Parsing | pdfplumber | Text extraction from resume PDFs |
| Validation | Pydantic v2 | Request/response schema validation |
| Skill Extraction | Custom regex taxonomy | 500+ skills across 12 categories |
| Testing | Pytest (205 tests) | Unit, integration, and scenario tests |
| Deployment | Render | Cloud hosting with auto-deploy |
Analyze a resume PDF against a job description.
Request — multipart/form-data
| Field | Type | Required | Description |
|---|---|---|---|
resume |
File (PDF) | ✅ | Text-based PDF, max 5 MB |
job_description |
string | ✅ | Min 50 characters |
Response — 200 OK
| Field | Type | Description |
|---|---|---|
fit_score |
float | Hybrid score 0.0–1.0 |
fit_percentage |
float | fit_score × 100 |
tfidf_score |
float | Raw TF-IDF cosine similarity |
skill_score |
float | Skill match rate 0.0–1.0 |
verdict |
string | Strong Fit / Partial Fit / Weak Fit |
advice |
string | Hiring recommendation |
matched_skills |
list[str] | Skills in both resume and JD |
missing_skills |
list[str] | Skills in JD but not in resume |
extra_skills |
list[str] | Skills in resume not in JD |
resume_skill_count |
int | Total skills found in resume |
jd_skill_count |
int | Total skills found in JD |
resume_text_length |
int | Character count of extracted text |
Error Responses
| Status | Meaning |
|---|---|
400 |
Invalid PDF, corrupt file, or no extractable text |
413 |
File exceeds 5 MB limit |
422 |
Missing field or job description too short |
500 |
Unexpected server error |
All errors: { "detail": "Human-readable message" }
ResuFit detects 500+ skills across 12 categories:
| Category | Examples |
|---|---|
| Programming Languages | Python, JavaScript, TypeScript, Go, Rust, Java |
| Web Frameworks | FastAPI, Django, Flask, React, Next.js, Spring Boot |
| Databases | PostgreSQL, MongoDB, Redis, MySQL, Elasticsearch |
| Cloud & DevOps | AWS, Docker, Kubernetes, Terraform, GitHub Actions |
| Machine Learning & AI | scikit-learn, PyTorch, TensorFlow, HuggingFace, BERT |
| Data Science | pandas, numpy, Spark, Airflow, Tableau |
| APIs & Integration | REST, GraphQL, gRPC, Kafka, RabbitMQ |
| Testing | pytest, Jest, Cypress, TDD, Selenium |
| Version Control | Git, GitHub, GitLab, Agile, Scrum |
| Security | OAuth, JWT, TLS, OWASP |
| Mobile | React Native, Flutter, iOS, Android |
| Soft Skills | Leadership, Communication, Mentoring |
resufit/
├── app/
│ ├── main.py # FastAPI app — middleware + route registration
│ ├── config.py # Centralized settings (thresholds, limits)
│ ├── exceptions.py # Global exception handlers + logging config
│ ├── middleware/
│ │ └── logging.py # Request logging with unique request IDs
│ ├── routers/
│ │ └── analyze.py # POST /api/v1/analyze endpoint
│ ├── schemas/
│ │ └── analysis.py # Pydantic request/response models
│ └── services/
│ ├── analyzer.py # Main pipeline orchestrator
│ ├── pdf_parser.py # PDF text extraction (pdfplumber)
│ ├── skill_extractor.py # Regex-based skill matching engine
│ ├── skills_taxonomy.py # 500+ skills across 12 categories
│ └── scorer.py # TF-IDF vectorizer + cosine similarity
├── tests/ # 205 tests across 8 test files
├── conftest.py # Shared pytest fixtures
├── pytest.ini # Pytest configuration
└── requirements.txt
- Python 3.10+
# 1. Clone the repository
git clone https://github.com/YOUR_USERNAME/resufit.git
cd resufit
# 2. Create and activate virtual environment
python -m venv venv
source venv/bin/activate # Windows: venv\Scripts\activate
# 3. Install dependencies
pip install -r requirements.txt
# 4. Start the development server
uvicorn app.main:app --reloadVisit http://localhost:8000/docs for the interactive Swagger UI.
# Run all 205 tests
python -m pytest
# Run a specific module
python -m pytest tests/test_scorer.py -v
# Run only scenario/integration tests
python -m pytest -m integration
# Run with short tracebacks
python -m pytest --tb=shortWhy keyword matching instead of spaCy NER? spaCy's NER doesn't recognize "FastAPI" or "TF-IDF" — they're not in any general-purpose NLP vocabulary. A curated keyword taxonomy gives deterministic, explainable results. In hiring tools, every decision must be justifiable.
Why 60% weight on skills vs 40% on TF-IDF? For technical roles, specific tool matches are more predictive of fit than writing similarity. A resume saying "orchestrated containerized deployments" scores high on TF-IDF but might not mention Docker. Skills get higher weight for precision.
Why lookahead/lookbehind instead of \b word boundaries?
\b fails for skills ending in non-word characters like c++. (?<!\w) and (?!\w) work universally regardless of whether the skill starts or ends with a symbol.
Why a dataclass for AnalysisResult?
Typed fields, IDE autocomplete, and a clear contract. The Pydantic schema in Step 6 mirrors it exactly, adding validation before the response leaves the server.
MIT License — see LICENSE for details.
Engr. Hasnain Zainulabdin R&R Digital Solutions
Contact: 03126641281 | HasnainZainulabdin@gmail.com Website: https://hasnainzainulabdin.vercel.app/
Built as a portfolio project demonstrating AI/ML API development with Python and FastAPI.