// demos
Every demo is a real app built on Layer that reimplements nothing. Each composes shipped gateway features over a different corpus and makes the gateway's decisions visible in the UI. The status and latency below are measured from your browser, right now — the same habit the demos have of showing their work.
checking 7 live apps…
checking…
Everything together — an end-to-end storefront.
An indexing pipeline, image-native semantic search (CLIP), recommendations, facets, and observability in one app. KEDA scales workers from pipeline metrics; Karpenter scales the nodes underneath. The demo that exercises the most of Layer at once.
checking…
Clinical search, with a number.
Clinicians search by exact token (metformin 500mg) and by clinical picture (elderly woman with progressive dyspnea). chart routes both — and scores it against real relevance judgments, so the claims are measured, not asserted. A GPU Gemma cascade extracts clinical events per note.
checking…
Hybrid text fusion, proven with qrels.
Purely lexical — no embeddings, no GPU. One query fans out into a BM25 leg plus a fuzzy leg per token, fused by reciprocal rank fusion. SciFact ships qrels, so every search scores nDCG@10 and shows a fusion inspector.
checking…
Book search that shows its routing.
One search box, three routes. Type an author, a title, or a vibe; the gateway picks keyword, semantic, or a fused blend from the shape of the query, and shelf renders that decision as a badge with the reason.
checking…
All of Simple English Wikipedia, routed and embedded on CPU.
The routing hero at corpus scale: one Auto query routes each search to full-text, semantic, or a fused RRF blend, and the UI renders the gateway's routing echo beside every result. The semantic leg is Lattice — an ~8 MB int4 lookup-table model served in-process on the gateway CPU. No GPU anywhere in the write or query path.
checking…
Text-to-image search with zero GPUs.
Cross-modal search: type sunset over water, get sunsets. The schema is two lines — an image_url attribute with a local CLIP embedding profile. The gateway runs CLIP's image tower in-process on CPU at write time and the text tower at query time; the app contains no embedding, tokenizer, or image-preprocessing code.
checking…
A search & AI jobs board with earned facets.
Daily pipelines index first-party public job postings; a guided-decoded Gemma cascade extracts location, work mode, salary, sponsorship, and role class as filterable facets. Every extracted value must clear a confidence floor and quote exact evidence from the posting — missing or ambiguous evidence stays unstated rather than guessed.
Want the implementation detail behind each one? Read the demos page in the docs.