NLP · 2026 · complete
Smart Resume Screener
Local-first ATS that ranks resumes against a job description using Ollama embeddings and a five-signal scoring model. No API keys, no data leaves the machine.
Project demo
Specs
- API
- FastAPI, Swagger at /api/docs
- Embeddings
- nomic-embed-text (local)
- Reasoning
- llama3.2:1b (local)
- Weights
- 50 embed / 20 keyword / 15 exp / 10 edu / 5 bonus
- Skill ontology
- 59 skills, ~190 aliases
- Batch limit
- 20 resumes (MAX_RESUMES_PER_BATCH)
- Upload limit
- 10 MB per file
- Deploy
- docker compose up
Features
- -Semantic match scoring: cosine similarity on nomic-embed-text vectors
- -Missing-skill detection through a ~190-alias, 59-skill ontology
- -Batch PDF / DOCX / text parsing, up to 20 resumes per request
- -Plain-language strengths, weaknesses and recommendation from llama3.2:1b
- -SHA-256 keyed disk cache for embeddings; CSV / JSON export
Architecture & tech stack
- 01 InputPDF via pdfplumber, DOCX via python-docx, or raw text.
- 02 ParseRegex and ontology extraction of name, contact, skills, years of experience, education level.
- 03 EmbedJob description and each resume embedded locally, cached by SHA-256.
- 04 ScoreFinal = 0.50 embedding + 0.20 keyword + 0.15 experience + 0.10 education + 0.05 bonus.
- 05 Reasonllama3.2:1b writes summary, strengths, weaknesses, missing skills and fit level.
- 06 ServeREST endpoints for upload, run, results and export; frontend/index.html served by the backend.
PythonFastAPIOllamanomic-embed-textllama3.2:1bpdfplumberpython-docxDockerpytest
Known limitations
- !Runs on local CPU, so a batch takes noticeably longer than a hosted model would.
- !A 1B-parameter model gives short, sometimes generic reasoning text.
- !Scoring weights are hand-tuned, not learned from hiring outcomes.
Local development guide
1. Pull the local models
# Install Ollama first: ollama.com
$ ollama pull nomic-embed-text
$ ollama pull llama3.2:1b2. Get the code
$ git clone https://github.com/shalombebx/smart-resume-screener.git
$ cd smart-resume-screener3. Virtual environment and dependencies
$ python -m venv venv
$ source venv/bin/activate # Windows: venv\Scripts\activate
$ pip install -r requirements.txt4. Environment and folders
# Defaults point at Ollama on 127.0.0.1:11434; no API key needed
$ cp .env.example .env
$ mkdir -p data/uploads data/cache exports5. Run it
# UI at localhost:8000, API docs at /api/docs
$ python main.py
# or: docker compose up --build
# or: python demo.py (terminal ranking of 5 sample resumes)