HN Showcase
Back to browse
Getting Warmer – Daily word game scored by GloVe embedding similarity

Getting Warmer – Daily word game scored by GloVe embedding similarity

by frostadvisory·Feb 19, 2026·2 points·0 comments
Visit ProjectView on HN

AI Analysis

●●SolidCrowd PleaserCozy

Wordle meets word vectors: temperature scoring by semantic distance, not spelling.

Strengths
  • •GloVe embedding cosine similarity is genuinely novel scoring mechanic versus letter-position games
  • •Leaderboard tracks both speed and efficiency (guesses to win), creating two skill axes
  • •Daily rotation + global leaderboard drives daily engagement without combat fatigue
Weaknesses
  • •No persistent or open-source component—pure SaaS game with limited extensibility
  • •Relies on GloVe static embeddings; lacks modern LLM-based semantic distance refinement
Category
Gaming
Target Audience

Word game enthusiasts and fans of semantic/meaning-based puzzles

Similar To

Wordle · Semantle · Spelling Bee

Post Description

Each day there's a new secret word — you guess words and get a temperature score from -40° to 100° based on cosine similarity of GloVe 100d embeddings. The closer your guess is in meaning to the target, the warmer your score. Built with Nuxt 3, AWS Amplify, and ~400k word vectors stored as binary chunks in S3. Happy to answer questions about the implementation.

Similar Projects

AI/ML●●Solid

Image prompt game with multi-signal CLIP/HSV/HOG scoring

The scoring model is the real clever bit: mixing semantic CLIP checks with a separate CLIP-based faithfulness test plus HSV and HOG-lite signals makes the feedback much harder to game than a single metric. The product wraps that into a tight, addictive loop—daily challenges, speed/unlimited modes and leaderboards—so it actually teaches you to write better prompts; just be aware it still inherits CLIP blind spots and could be optimized around the exact weighting scheme.

Big BrainSlickCrowd Pleaser
gosu94
114mo ago