Solwend
Solwend is a wayfinding platform that routes walks on felt-minutes rather than shortest distance, modeling sun, heat, canopy, elevation, and scenic quality. As of May 2026 it's a substrate owning identity, map, places, and a taste graph.
Solwend is a felt-minute wayfinding platform in the General Systems Ventures portfolio, filed at pace-layer J (Geology / Place) as GSV-J1. The core routing engine is runnable Rust/Python, not a spec: for a Downtown Los Angeles Phase-1 spike it pulls an OpenStreetMap walking graph, derives a digital surface model from USGS LIDAR, ray-traces a shadow raster at a summer-afternoon sun position, scores each street segment by percent-shaded, and runs A* over felt-minute edge weights, where an uncomfortable minute costs more than a comfortable one. In May 2026 it pivoted from a single app to a platform substrate on the HealthKit model: the substrate owns spatial truth (map graph, places, paths, observations), identity ('Sign in with Solwend'), and a cross-app derived-taste graph, while consumer apps own only persona/voice/UX and talk to it over typed-event HTTP — emitting typed facts like `walk_completed`, never interpretations like `prefers_shade=true`, so the substrate accumulates ground truth and derives taste itself. Monorepo layout: `platform/` (spatial + identity + graph + a canonical destination schema in Neon Postgres) alongside three consumer prototypes in `apps/` (stepsdaddy, uncrancher, fashionsomething) running on mock data. The wikifeedia destination schema is live with several thousand imported restaurants plus citations and awards; the identity provider and graph ingest are still stubs. Methodology is borrowed openly from published comfort science — UTCI via pythermalcomfort, OSM, and the CoolWalks/ThermalScape shadow-routing literature. Related portfolio projects: GSV-J1b (an undisclosed DIY virtual-sky sibling build), RaveNous (GSV-I1, which began as a Solwend consumer app before being cut to its own clock), and jawnfit (which reuses Solwend's sun-vector shadow ray tracer for indoor scenes). Honest status as of mid-2026: pre-product — a working spatial spike proves the divergence from shortest-path routing, the destination database is populated, and persona prototypes iterate against mock data while the substrate is built bottom-up; no live deployment, no public app, no SDK yet.