An intelligent CV tailoring system that optimizes resumes for specific job postings using advanced LLM technology and cosine similarity analysis.
⚠️ ALPHA VERSION WARNING⚠️ This entire project is currently in ALPHA stage and should be considered experimental. Both the FastAPI backend and Streamlit GUI are under active development and not recommended for production use. Use at your own risk for testing and experimentation purposes only.
Knitty is a production-ready AI-powered CV tailoring system that optimizes resumes for specific job postings. The system analyzes your existing CV alongside a target job posting, then generates an optimized version that better aligns with the job requirements while maintaining authenticity and professional quality.
You can access Knitty through:
- ⚙️ FastAPI Backend (
⚠️ ALPHA): RESTful API for programmatic access and integration - 🖥️ Streamlit GUI (
⚠️ ALPHA): User-friendly web interface for interactive CV enhancement
⚠️ Both interfaces are experimental and in active development.
- 🔄 Multi-LLM Architecture: Leverages four specialized AI models for different tasks
- 📄 PDF Processing: Extracts and processes CV content from PDF files
- 🌐 Web Scraping: Automatically fetches job postings from LinkedIn and other platforms
- 🧠 RAG-Powered Analysis: Uses Retrieval-Augmented Generation for intelligent job content extraction
- 📊 Similarity Scoring: Measures CV-job alignment using cosine similarity
- 🎯 Smart Enhancement: Generates tailored CVs while maintaining professional quality
- 🔁 Iterative Improvement: Automatically retries generation if similarity doesn't improve
- 📱 HTML Export: Converts final CVs to professional HTML format
The system operates through a sophisticated multi-stage pipeline:
---
config:
theme: redux
---
flowchart TD
A["CV PDF"] --> B["PDF Text Extraction"]
A1["Additional Info"] --> B1["Combine CV Content"]
B --> B1
B1 --> C["CV Keyword Extraction<br>Fast LLM"] & D["CV Embedding Generation<br>Embed LLM"] & M["Smart LLM Enhancement"]
E["Job URL/Text"] --> F{"Input Type?"}
F -- URL --> G["Web Scraping<br>Playwright + BeautifulSoup"]
F -- Text --> H["Direct Text Processing"]
G --> I["RAG Processing<br>Context LLM + Vector Store"]
H --> I
I --> J["Job Keyword Extraction<br>Fast LLM"] & K["Job Embedding Generation<br>Embed LLM"] & M
C --> L["Cosine Similarity Analysis"] & M
D --> L
J --> L & M
K --> L
L --> M
M --> N["Enhanced CV Generation"]
N --> O["Similarity Check"]
O -- Improved --> P["Final CV Output"]
O -- Not Improved --> Q["Iterative Retry"]
Q --> M
P --> R["HTML Export<br>Special Sauce API"] & S["Markdown Export"]
A:::Ash
M:::Ash
E:::Ash
L:::Ash
P:::Ash
classDef Ash stroke-width:1px, stroke-dasharray:none, stroke:#999999, fill:#EEEEEE, color:#000000
style A fill:#e1f5fe
style M fill:#f3e5f5
style E fill:#e8f5e8
style L fill:#fff3e0
style P fill:#e8f5e8
- Fast LLM (for example: GPT-OSS-120B): Quick keyword extraction from CV and job postings
- Context LLM: Large-context processing for web-scraped content and RAG operations
- Embed LLM: Generates high-quality embeddings for similarity analysis
- Smart LLM (Currently using: Gemini 2.5 Pro): Advanced CV enhancement and generation
- LangChain: Document processing, embeddings, and LLM orchestration
- Playwright: Robust web scraping for job postings
- BeautifulSoup: HTML cleaning and text extraction
- NumPy: Mathematical operations for similarity calculations
- PyPDF: PDF text extraction and processing
- RAG Pipeline: Uses InMemoryVectorStore for intelligent content retrieval
- Async Processing: Non-blocking web scraping with proper Windows compatibility
- Template System: YAML-based CV templates for consistent formatting
- Configuration Management: File-based prompt and template management
- Error Handling: Comprehensive error recovery and retry mechanisms
# Clone the repository
git clone https://github.com/Creative-Geek/Knitty.git
cd Knitty
# Install dependencies
pip install -e .
# Install Playwright browsers for web scraping
playwright install chromiumCreate a .env file in the project root with your LLM API configurations:
# Fast LLM (Keyword Extraction) - e.g., Groq Llama 3
FAST_LLM_API_KEY="your-api-key"
FAST_LLM_API_BASE="https://api.groq.com/openai/v1"
FAST_LLM_MODEL_NAME="llama3-8b-8192"
# Smart LLM (CV Enhancement) - e.g., OpenAI GPT or Google Gemini
SMART_LLM_API_KEY="your-api-key"
SMART_LLM_API_BASE="https://api.openai.com/v1"
SMART_LLM_MODEL_NAME="gpt-4-turbo"
# Embedding LLM (Similarity Analysis) - e.g., OpenAI Embeddings
EMBED_LLM_API_KEY="your-api-key"
EMBED_LLM_API_BASE="https://api.openai.com/v1"
EMBED_LLM_MODEL_NAME="text-embedding-ada-002"
# Optional: HTML Generation Service
SPECIAL_SAUCE_API_URL="https://cv-service.com/generate"
SPECIAL_SAUCE_API_KEY="your-service-key"
⚠️ ALPHA WARNING: Both interfaces are experimental. Use for testing and development only.
python app.py
# or
uvicorn app:app --host 0.0.0.0 --port 8000 --reloadAccess the API at http://localhost:8000 with interactive API docs at http://localhost:8000/docs
streamlit run streamlit_app.pyOpen your browser to http://localhost:8501 and start enhancing CVs through an intuitive interface.
jupyter notebook notebook.ipynbGreat for experimentation and research.
- Extracts text from PDF CVs using PyPDFLoader
- Combines CV content with additional information
- Processes job postings via web scraping or direct text input
- Uses Fast LLM to extract professional keywords from both sources
- Returns structured JSON arrays of relevant terms and phrases
- Focuses on technical skills, experience, and job requirements
- Generates embeddings for keyword sets using specialized models
- Calculates cosine similarity to quantify CV-job alignment
- Establishes baseline metrics for improvement tracking
- Uses Smart LLM to generate optimized CV based on analysis
- Maintains professional tone and template compliance
- Includes iterative improvement with similarity feedback
- Measures improvement in cosine similarity scores
- Automatically retries if enhancement doesn't improve alignment
- Exports final CV in both markdown and HTML formats
The system typically achieves:
- Baseline Similarity: 40-70% initial CV-job alignment
- Improvement Range: +5% to +15% similarity increase
- Quality Metrics: Professional presentation with strategic keyword integration
- Processing Time: ~30-60 seconds for complete enhancement
You'll have to wait for that 😉
We welcome contributions to enhance and evolve Knitty! Areas where we could use help:
- Frontend Enhancement: Improve the Streamlit GUI with additional features
- Backend Optimization: Performance improvements to the FastAPI server
- Model Fine-tuning: Optimize LLM prompts and enhance CV generation quality
- DevOps: Docker containerization and cloud deployment setup
- Testing: Expand test coverage and add integration tests
- Documentation: API documentation and user guides
- Feature Development: New features like batch processing, templates, and analytics
- Architecture Guide - Detailed system architecture
- API Reference - Function signatures and data models
- Development Guide - Setup and development instructions
- Scoring Process - Similarity calculation methodology
This project is currently in research and development phase. License terms will be established as the project evolves toward production release.
- LangChain team for excellent LLM orchestration tools
- OpenAI for embedding and language model APIs
- Playwright team for robust web automation capabilities
- The open-source community for inspiration and foundational technologies
Ready to revolutionize your job applications? Get started with Knitty today!