Pyraword

Inspiration

We were inspired by the daily challenge and replayability of Wordle, as well as the viral word-association puzzles shared across social media. We wanted to create a game that combines vocabulary knowledge with logical deduction while capturing the qualities of an iconic daily-playable game.

What it does

The goal of the game is to complete a word pyramid by guessing the missing words. Each word is semantically related to the previous one, and a definition is provided to help guide the player. A new sequence of words is generated each day, ensuring fresh puzzles and replayability.

How we built it

We used the Devvit template for Reddit integration and Python for puzzle generation and data handling.

No artificial intelligence or machine learning models were used in the puzzle generation logic, only for creating visual assets (logo and background).

The puzzles are generated using the Open English Wordnet database, derived from Princeton WordNet and maintained by an active community. We parse the XML dataset and construct chains of semantically related words by navigating synonym relationships between word senses.

Challenges we ran into

A major challenge was determining meaningful relationships within WordNet. The dataset is extremely large and densely connected, so naïve traversal either produced unsolvable puzzles or trivial ones. We had to design search constraints and heuristics (such as weighted starting words and guided depth-first search) to reliably generate valid and interesting word chains.

Accomplishments that we're proud of

We successfully built a fully functional daily word puzzle that automatically generates solvable sequences without relying on AI models.

What we learned

We learned how to parse and process large XML linguistic datasets in Python, how lexical databases like WordNet are structured, and how to integrate a backend generator with the Devvit framework, along with discovering some interesting new words along the way.