Branded intelligence.

For oncology, rare disease, and immunology. Set per brand. Built to learn. Smarter every quarter.

OVYN builds one intelligence per brand. Each shell learns its brand’s voice, rules, claims library, and the sources that move its market. Marketing, content, and ops run on top.

Four shells live on the platform today. Oncology HCP, rare-disease (incognito), compounding wholesale, and the home brand that built it. Same brain architecture underneath. Audited by default.

Launched 2026-05  ·  in trial with an independent network  ·  4 shells live
The shells  ·  in production

Lumen Oncology

Pharma · oncology · HCP

VANATAR launch under MLR review. HCP educational content + formulary submission materials. Brand-locked to VANTAGE trial citations, fair-balance pairing structural, not bolted on.

Product X

Pharma · rare disease (anonymized)

Rare-disease brand onboarded from a single brand-book PDF in 30 minutes. Caregiver + HCP carousels generated against the brain in 36 seconds each, every execution cites a claim by ID, rails catch what MLR shouldn't have to. Real client, incognito for external demo.

Nourish Health

Pharma · compounding · wholesale

Compounded-medication distributor. Brand-locked distributor-to-clinic collateral with three-axis compliance (rules · self-check · brand-fit): claims library, forbidden phrases, MLR-safe educational templates that clinics localize.

Ephicacy Health

Pharma · regulated content · home brand

The home brand that built OVYN. Brand brain trained on a claims library, voice rules, and MLR-derived tonal preferences. Reference shell for every new tenant onboarded onto the platform.

Products built on OVYN, sold into these verticals, operate at the shells’ URLs above.
How a shell is built
  1. Brand essence anchored. Thesis, story, truths, anti-positioning, tells. This is the spine the model never crosses.
  2. Brand corpus ingested. 50–500+ pieces of canonical brand writing. Semantic retrieval at inference time so every generation is grounded in the brand’s actual voice.
  3. Three-axis check on every output. Model self-check, deterministic phrase scan, and brand-fit identity scoring. Brand-fit is independent: a piece can pass compliance but fail on voice, or vice versa.
  4. Approved executions auto-promote into the brand's precedent library. The adaptive per-brand model retrieves top-K precedents at inference time as match-this-voice anchors. Sharpens with every approval. Every precedent used is citable in the audit drawer.
  5. Full provenance on every piece. Brain version, PromptGraph, three-axis trace, brand-fit score. Logged. Auditable.