AI-powered resume intelligence app with two practical modes:
JD Matchmode for resume-vs-job fit scoringResume Healthmode for standalone resume quality scoring and fixes
It is production-ready for portfolio use with Streamlit UI, FastAPI backend, CI tests, evaluation artifacts, Docker support, and Hugging Face auto-deploy.
- Hugging Face Space: https://huggingface.co/spaces/shinzobolte/ai-resume-screener-job-matcher
- GitHub Repository: https://github.com/shinzoxD/ai-resume-screener-job-matcher
Recruiters and candidates waste time on manual resume screening because most tools are either:
- keyword-only and shallow,
- hard to explain,
- not ATS-aware,
- or not actionable for resume improvement.
ResumePilot AI solves this with a clear, auditable workflow:
- Extract text from resume and JD PDFs/text.
- Compute hybrid match scores (semantic + lexical + skill alignment).
- Show matched/missing skills and ATS indicators.
- Generate concrete improvement suggestions (LLM when configured, rule-based fallback otherwise).
- Provide downloadable outputs for next-step resume edits.
JD Match
- Inputs: Resume + Job Description
- Outputs: Overall match, ATS score, skill alignment, confidence, missing skills, targeted suggestions
Resume Health
- Inputs: Resume only
- Outputs: Resume quality score, section quality, contact completeness, bullet quality, quantified impact, top fixes
- PDF/Text input support (resume and JD)
- Hybrid similarity scoring with
sentence-transformers+ cosine + lexical features - Skill extraction with matching/missing skills view
- ATS compatibility scoring and keyword density checks
- Interactive score card UI with ring score and issue breakdown
- LLM suggestions via Groq (
GROQ_API_KEY) with safe fallback to rule-based suggestions - Batch screening mode
- FastAPI analyze endpoint for backend integration
- Download outputs (PDF/Markdown)
- CI-tested + Dockerized + Hugging Face auto-deploy
- UI:
streamlit - PDF parsing:
PyMuPDF (fitz) - Embeddings:
sentence-transformers(all-MiniLM-L6-v2,all-mpnet-base-v2) - ML metrics/utilities:
scikit-learn,numpy,scipy,pandas - Optional LLM:
groq(Llama / Mixtral) - Backend API:
fastapi+uvicorn - Testing:
pytest - Deployment: Docker + GitHub Actions + Hugging Face Spaces
- Python:
3.11+
flowchart TD
A["Resume PDF or text"] --> B["Extraction and cleaning"]
A2["JD PDF or text"] --> B
B --> C["PII redaction optional"]
C --> D["Scoring engine"]
D --> D1["Semantic similarity"]
D --> D2["Lexical overlap"]
D --> D3["Skill alignment"]
D --> D4["Confidence calibration"]
D --> E["ATS and keyword analytics"]
D --> F["Explainability and strengths"]
F --> G["Suggestions engine"]
G --> G1["Groq LLM when key exists"]
G --> G2["Rule based fallback"]
E --> H["Streamlit result dashboard"]
F --> H
G --> H
D --> I["FastAPI analyze endpoint"]
No demo video included (as requested). Static UI snapshots are included below.
.
|-- app.py
|-- backend/
| |-- __init__.py
| |-- main.py
| `-- schemas.py
|-- utils/
| |-- extractor.py
| |-- matcher.py
| |-- llm_suggestions.py
| |-- observability.py
| |-- privacy.py
| `-- skills_db.py
|-- data/
| |-- app_history.db
| `-- eval_pairs.jsonl
|-- scripts/
| |-- evaluate.py
| `-- capture_screenshots.py
|-- tests/
| |-- conftest.py
| |-- test_api.py
| |-- test_extractor.py
| |-- test_matcher.py
| |-- test_new_features.py
| `-- test_privacy.py
|-- artifacts/
| `-- eval_results.csv
|-- assets/
| `-- screenshots/
|-- .github/workflows/
| |-- ci.yml
| `-- deploy-hf.yml
|-- .streamlit/config.toml
|-- requirements.txt
|-- pyproject.toml
|-- Dockerfile
|-- docker-compose.yml
|-- Makefile
`-- README.md
As of March 2, 2026, from local project artifacts and tests:
- MAE:
16.711 - RMSE:
19.320 - Pearson correlation:
0.949 - Spearman correlation:
0.818
- Test suite:
12 passed(pytest -q) - Python compile check:
app.pycompiles successfully - CI pipelines included: unit tests + Hugging Face deploy workflow
python -m venv .venv
.venv\Scripts\Activate.ps1
pip install --upgrade pip
pip install -r requirements.txt
streamlit run app.pyOptional API server:
uvicorn backend.main:app --reload --host 0.0.0.0 --port 8000Run tests:
pytest -qGROQ_API_KEY = "your_groq_key"Add GROQ_API_KEY in Space Settings -> Variables and secrets.
- App still works fully.
- Suggestions automatically switch to rule-based mode.
- UI shows a warning/notice so users know LLM suggestions are disabled.
- Push this repo to GitHub.
- Go to https://share.streamlit.io.
- Select repo and
app.pyas entrypoint. - Add optional secret
GROQ_API_KEY.
This repo includes .github/workflows/deploy-hf.yml.
Required GitHub secrets:
HF_TOKEN(write token)HF_SPACE_REPOinowner/space-nameformat
On push to main, workflow syncs repo to your Docker Space.
- Built and deployed
ResumePilot AI, a dual-mode resume intelligence platform (JD Match + Resume Health) using Streamlit, FastAPI, PyMuPDF, and sentence-transformers, with ATS analytics and explainable scoring. - Engineered a hybrid scoring pipeline combining semantic similarity, lexical overlap, and skill alignment with confidence calibration; achieved strong alignment to human labels (Pearson
0.949, Spearman0.818) on evaluation pairs. - Implemented production CI/CD with pytest-based quality gates and automated GitHub-to-Hugging-Face deployment, plus optional Groq LLM suggestions with resilient rule-based fallback.
- Add role-specific score calibration datasets.
- Add richer grammar/style checks for Resume Health mode.
- Add recruiter workspace with persistent candidate comparison.
- Add benchmark suite for multilingual and OCR-heavy resumes.
- Add optional vector database for historical profile search.


