Allivista
Your research ally with a view.
Structural Intelligence/GeoLang/Allivista

Allivista is the working proof of concept for Structural Intelligence — a structural layer that works alongside AI, not in place of it. It surfaces structurally-coherent connections in public research corpora that a field's dominant prose narrative hasn't yet stated; a model can then narrate those connections in plain language, never generate them. The discovery is pre-cognitive and deterministic — no LLM in the analysis path, no hallucination surface in the structure — with the raw result shown alongside as receipts the narration can be checked against.

First Pillar · Alzheimer's research, June 2026
From 9,858 PubMed Alzheimer's abstracts, Allivista surfaces a structurally-coherent cluster of cardiovascular, vessel, cerebrovascular, heart, and kidney concepts as zero-direct-co-occurrence neighbors of the Alzheimer anchor — the classic Don-Swanson hidden-bridging pattern, produced from pure graph structure.

A clinically-informed reading of this cluster inverts the brain-anchored framing of Alzheimer's disease into a systemic-clearance-failure phenotype, of which brain pathology is the visible end-state. The interpretation is consistent with the emerging Klotho, neurovascular-unit, glymphatic, ARIA, AMBAR, and CKD-dementia literatures. The system did not generate the hypothesis. It surfaced the structural fingerprint a clinician then recognized.
9,858 abstracts · 200,685 symbols · 1.92M edges · 45.82s training · 18ms synthesis
Structural path vs. raw context · same model, same question
We posed one research question to the same model two ways: once with Allivista's structural tools, once with raw PubMed papers loaded into context. The structural path used ~2.9× fewer input tokens and less than half the cost, and returned organ-system-level findings robust to which papers were sampled. The raw-context path could fit fewer than 1% of the corpus and failed outright past the model's token limit — beyond ~100 papers, structural distillation is not an optimization but the only path that runs.
57,006 tokens structural vs 164,952 raw · 2.89× ratio · the cost

The full Swagger interface lets you call any endpoint from the browser. The canonical Alzheimer's space is loaded and ready — try a Swanson-style ABC discovery to reproduce the cluster above, or run the 5-method convergence analysis to see how it partitions the structural space.

# Don-Swanson ABC discovery, random-walk weighting curl -X POST https://api.allivista.com/v1/synthesize/swanson \ -H "Content-Type: application/json" \ -d '{ "space": "_alzheimer_space", "anchor": ["alzheimer","alzheimer'\''s","alzheimers","eoad","fad","sad"], "weighting": "random_walk", "rank_by": "discovery_ratio", "top_n": 10 }'

Each event is retrievable at https://api.allivista.com/v1/evidence/{event_id}. Every event is signed by content-derived SHA-256 in a parent-event chain; no event can be retroactively altered without invalidating every later event's hash. The full append-only chain is available on request.

Patent and trademark applications in process. The GeoLang substrate is the patent target; Allivista is the trademark target. Mechanism-level architecture details (placement, scoring, bond extraction, neighborhood patterns) are held privately pending utility patent disclosure. The evidence chain anchors the timeline of every architectural decision under content-derived SHA-256 hashing.

contact@allivista.com
Substantive academic and institutional inquiry welcome. Press and licensing discussion under separate cover.