Modex engine
Modex is an experimental machine learning platform for discovering hidden structure in complex systems — built on ingestion, embeddings, similarity scoring, clustering, and hypothesis summaries.
Composable stack
Next.js routes call thin orchestration in lib/modex-core.ts, backed by engine/embedding, engine/similarity, and patterns/*. Replace the deterministic embedding with hosted models when keys and governance are in place.
Typed text ingest
Experiments accept structured descriptions today — the architecture anticipates audio and time-series adapters without changing the orchestration contract.
Embeddings & similarity
Signals map into vectors for pairwise comparison; scored neighbors feed both UI graphs and cluster formation.
Clustering & hypotheses
Groups are formed by similarity thresholds; each cluster exposes an early hypothesis tier you can validate with domain experiments — not a verdict.
Supabase (optional)
When NEXT_PUBLIC_SUPABASE_* env vars are set, signals persist and similarity pulls from your dataset; otherwise the console runs in local discovery mode.