How rx flows across 4 doors
A single operator runs four life domains — fitness, nutrition, finance, learning — each backed by its own data pipeline, knowledge corpus, and storage backend. The rx (prescription) layer is the unified recommendation engine that reads each domain's live signals, consults a personal knowledge vault, and proposes one ranked, sourced, counter-thesised action at a time.
This explainer is visual first. Start with the overview, then follow the path that matches your role.
System architecture
Hover any node to see what it's connected to. Drag nodes to rearrange. Use the controls bottom-left to zoom, fit, or lock the view.
Four doors, each with its own storage: Zeus (fitness) → Supabase, surfaced in the Lovable verocity app · Athena (nutrition) → Supabase, no UI consumer · Lakshmi (finance) → TradingV local postgres, surfaced in a FastAPI panel · Ganesh (learning) → markdown only. Every door consults the shared knowledge vault for evidence; no door writes to another door's backend.
Four doors, each with its own storage: Zeus (fitness) → Supabase, surfaced in the Lovable verocity app · Athena (nutrition) → Supabase, no UI consumer · Lakshmi (finance) → TradingV local postgres, surfaced in a FastAPI panel · Ganesh (learning) → markdown only. Every door consults the shared knowledge vault for evidence; no door writes to another door's backend.
Pick a path
Overview
What rx is. What the 4 doors mean. Why the storage routing is split. Read this first.
Ingestion
Every data source that flows in — markets, screenshots, videos, workouts, filings. Cadence and destination for each.
rx pipeline
How a single recommendation is born. Eleven numbered steps plus two reconcile pre-steps, counter-thesis rule, auto-dedup, snooze + auto-revive.
The 4 doors deep dive
Per-door storage, signals math, UI surface, status. Drift composite weights spelled out.
Feedback loop
What happens when you act, dismiss, or snooze. Phase W reconciler. Action-rate as system-death signal.
Deep tech
Embedding model choice, chunking, the four-layer hybrid retrieval, graph re-ranking math.
Maintenance
Rot vectors. The 6 touch points to add a new door. Hardcoded coupling to find before refactoring.
What this isn't
Scope: information architecture only. This explainer maps how data moves and lives. It does not cover the UI design of the verocity or TradingV apps, the trading strategy logic, or the fitness programming methodology. Those concerns have their own docs.
AGENTS.md for the re-verify workflow when source repos change.