Alex Roessner 罗轩阳
selected work · MMXXVI
← selected work

Orbital

Tensor-network architecture for real-world-asset corridors.

Operational; deal-graph and corridor-intelligence calls live. TypeScript 842 KB active 6h ago

Origin

Real-world-asset tokenization treats every deal as a one-off. A new sub-sovereign issuer, a new legal opinion, a new compliance memo — every time. The same playbook gets re-derived under a different deal name and stored in a different folder. The intelligence is in the corridor — the structural pattern that recurs — but the data model is the deal.

Problem

Without a representation that lets corridors surface as first-class objects, you cannot answer the questions that matter: 'has this issuer class been done before, where, with what method, with what outcome?' is supposed to be a millisecond query. In practice it is a week of partner-time and a slack thread.

Approach

A tensor representation in which deals, jurisdictions, methods, receipts, and outcomes are facets of a single high-dimensional structure. Cross-matrix queries surface 'same playbook, different deal' instantly: jurisdiction A × method B × outcome C is a coordinate, not a search. Built on Postgres + pgvector with a custom MCP layer that lets multiple assistants reason against the same tensor.

Methodology

Postgres-first. Every entity (deal, issuer, jurisdiction, regulation, opinion, receipt) is a row with explicit foreign keys; the tensor is computed over those rows on demand rather than stored as a blob. The MCP layer exposes specific introspection calls — corridor_intelligence, deal_graph, similar_deals, regulatory_diff — so that an assistant can navigate the tensor without ever seeing raw rows.

Selected milestones

Open questions

Ask me about

Stack

PostgreSQLpgvectorTypeScriptMCPSupabase
To dig in — alex.roessner@landseed.earth