Experiments that run in your browser.
Original tools at the seam of data architecture and AI — knowledge graphs, data-quality engines, and on-device models. The compute runs in your browser; nothing you load is uploaded — bar one demo that reads a live public transport feed. Built by John Mikel Regida.
The Edge Device
open ↗A 3D phone, built in three.js, that is a whole data platform — generate, profile, query with DuckDB, search by meaning, build a knowledge graph and run an in-browser LLM. One dataset flows through every app.
Ontology Explorer
open ↗A national dataset re-expressed as an interactive 3D knowledge graph — classes, instances, relations you can orbit and search.
Data Copilot
open ↗A real LLM running fully in your browser via WebGPU — ask questions of a dataset or get a data-quality read, nothing leaves the device.
Semantic Lint
open ↗On-device embeddings catch the errors rules can't — near-duplicates, semantic outliers, mislabels — with a 2D meaning map.
Data-Quality Hub
open ↗Three embedding tools in one: infer a schema's meaning, measure semantic drift between snapshots, and flag breaking schema changes.
MCP Tool Forge
open ↗Lints an MCP / agent tool schema the way an LLM reads it, previews the tool card, and shows a sample agent call. Data contracts for agents.
NaPTAN Explorer
open ↗All 435,029 UK transport stops, queryable in-browser with DuckDB-WASM — freshness, validity and completeness over the full dataset.
Data Profiler
open ↗A data-quality engine written in Zig, compiled to WebAssembly — drop in a CSV/TSV and profile types, nulls, cardinality and duplicates at ~1M rows/s.
TfL Live Status
open ↗Real-time London line status — Underground, Overground, Elizabeth line, DLR and trams — fetched live in-browser from the TfL Unified API.