Client Side AI Experiments

Browser-native AI demos powered by JavaScript and WebAssembly. No servers, no API keys, no data leaves your device.

Sub-millisecond nearest-neighbor search over 1,000 vectors using HNSW graph traversal in a 42 KB WASM microkernel.

Semantic emoji search — type a feeling, concept or word and find matching emojis by meaning, not exact keywords. 130+ emojis indexed in-browser.

WASM feature hashing cosine similarity

🏷️ Dansk Navnegenkendelse (NER)

Find personer, organisationer og steder i dansk tekst. En dansk ModernBERT-model (~144 MB) kører direkte i browseren via ONNX Runtime — ingen data sendes til en server.

Transformers.js ONNX BERT dansk NER

🎬 Dansk Filmsøgning

Semantisk søgning i danske film — skriv en beskrivelse og find film der matcher betydningen. Flersproget AI-model (~33 MB) kører direkte i browseren via ONNX.

Transformers.js ONNX gte-small semantic search dansk

🧠 RuvLLM WASM

Browser-native LLM toolkit — chat template formatting (Llama3, Mistral, ChatML, Phi, Gemma), HNSW semantic routing, and MicroLoRA adaptation, all running in ~150 KB of WebAssembly.

WASM LLM HNSW chat templates semantic routing

🪨 Rock Paper Scissors (N-gram vs SONA)

Play against two competing AI predictors: an n-gram frequency counter and a SONA adaptive learner (227 KB WASM). Both predict your next move independently — live charts and per-model metrics let you compare how they learn differently.

WASM SONA n-gram k-NN LoRA game AI