A FastAPI-powered proof-of-concept (PoC) application that parses resume data from PDF, DOCX, images, or LinkedIn URLs using a hybrid of OpenAI's GPT models and LandingAI's OCR API.
-
PDF / DOCX Resumes
- Parsed with PyMuPDF and
python-docx - Cleaned and structured using OpenAI GPT
- Returns key fields with confidence scores
- Parsed with PyMuPDF and
-
LinkedIn URLs
- Scraped using headless Selenium
- Parsed using GPT
⚠️ Subject to scraping limitations (bot protection, dynamic content)- Known issue: LLM may hallucinate/fill missing data (e.g., adding nonexistent roles)
-
Image Resumes (PNG / JPG / JPEG)
- Uploaded as single or multiple images
- Parsed using LandingAI’s Resume Parser API
- Ideal for scanned or photographed resumes
| Method | Endpoint | Description |
|---|---|---|
| POST | /parse |
Upload and parse a PDF or DOCX file |
| POST | /parse/Linkedin_url |
Submit a LinkedIn profile URL for scraping |
| POST | /parse/images |
Upload 1+ image files to parse with LandingAI |
| GET | / |
Health check for server |
Start your local FastAPI server with:
uvicorn main2:app --reloadVisit Swagger UI to test: 📍 http://127.0.0.1:8000/docs
Create a .env file in the root directory with your API keys:
OPENAI_API_KEY=your_openai_key
VISION_AGENT_API_KEY=your_landingai_keyInstall dependencies and run the server:
pip install -r requirements.txt
uvicorn main2:app --reload- FastAPI – lightweight API framework
- OpenAI GPT – resume field extraction from text
- LandingAI – resume OCR via Vision Agent
- LinkedIn scraping may result in incomplete data or fail due to privacy restrictions
- LLM hallucination possible when input is unstructured or minimal
- LandingAI API requires valid key and resumes must be image files only
- Handle multilingual or low-data resumes
- Add fallback test pages for LinkedIn if scraping fails
- Explore Claude or other LLMs for richer comparisons
- Add user interface for manual corrections on outputs
Built by Hitomi Hoshino 💡 Focused on resume intelligence, OCR evaluation, and hybrid AI pipelines.