Would like to see explanations for the connections which are sometimes not obvious.
Maybe make a video showing how it adapts as you make guesses. I wasn't sure what was happening when the first green arrow showed up.
I also like NYT Connections and have the feeling a better version could be created that would, like this, let you try out 2 and 3-word groupings visually before committing to a possible red-herring.
Originally designed to map world history, Constellations has evolved into a Universal Bipartite Explorer. It uses LLMs to identify the fundamental "Atomic" building blocks and the "Composite" collections that connect them in any domain.
Supported Bipartite Pairs:
History: Person (Atomic) ↔ Event/Project (Composite)
Cinema: Actor/Director (Atomic) ↔ Movie/TV Show (Composite)
Sports: Player (Atomic) ↔ Team (Composite)
Culinary: Ingredient (Atomic) ↔ Recipe (Composite)
Medicine: Symptom (Atomic) ↔ Disease (Composite)
Academia: Researcher (Atomic) ↔ Paper/Grant (Composite)
I built this to see if I could create 'collaboration graphs' on the fly using LLMs instead of a pre-computed database. It connects people to people through events they both participated in.
The project was 100% 'vibe coded'. I started in Google AI Studio, moved to Antigravity, then Cursor, and finally Codex. The README has the full 'tool-hopping' story and the initial prompt that started it all.
Link to the app: https://show-hn-classified.vercel.app/