A minimal semantic word-guessing game built with Streamlit. Challenge yourself to find the secret word by exploring semantic relationships.
- Semantic Similarity: Feedback based on cosine similarity scores.
- Intelligent Hints: Reveal nearby words to help you narrow down the target.
- Reveal Word: Stuck? You can reveal the secret word to see how close you were.
- Clean Interface: A distraction-free, typography-focused design.
- Fast Performance: Powered by Pinecone vector search for instant similarity scoring.
- Logic: Python, NumPy, Scikit-learn
- Database: Pinecone (Vector database for similarity search)
- Environment: Python-dotenv
- Data: ConceptNet Numberbatch word embeddings
-
Clone the repository:
git clone https://github.com/kavin-jindal/SemWord.git
-
Install dependencies:
pip install -r requirements.txt
-
Environment Setup: Create a
.envfile in the root directory and add your Pinecone API key:PINECONE=your_api_key_here
Note: An index named
semword-indexis required in your Pinecone project. -
Run the application:
streamlit run app.py
I was experimenting with Vector Embeddings and Pinecone DB a week ago when the idea of building a semantic based Wordle variant struck me. There are already such games available yet I undertook this project solely as a learning experience. Over the week long development cycle, I spent most of my time experimenting with different word embeddings and models starting from HuggingFace's all-MiniLM-L6-v2 to GloVe, Google-News-300, Word2Vec, Cohere's embeddings and finally settling on ConceptNet Numberbatch for its balance of performance and size. I also had to clean the embeddings to remove unecessary words and phrases considering the file was pretty big and would take too much time to load.
I manually wrote the backend after a lot of experimentation and tinkering and setup a working prototype on Streamlit. After that I used Claude Code to improve the UI because I absolutely hate doing frontend design myself.
In order to speed up the backend and handle search efficiently, I integrated Pinecone DB. This allows for extremely fast similarity scoring and vector search directly in the cloud, removing the need for heavy local embedding files.
I hope you like the project, any contributions to improve the project are welcome.