Platform

The platform under every Roshan AI product.

One clinical core turns ingestion, model intelligence, grounded reasoning, and APIs into products that can compound without a re-architecture.

Architecture map

Four layers. One clinical contract.

Each layer remains reusable as products, workflows, and integrations grow.

Layer 01
Ingestion
Clinical data, normalized into something models can read.

Notes arrive through API or batch, are normalized under customer policy, and become the canonical clinical representation downstream models consume.

Free-text notes and discharge summariesFHIR-flavored EHR exportsConfigurable PHI taggingPer-tenant data isolation
Layer 02
Models
Clinical encoders trained on real clinical text.

The encoder is paired with a learnable concept-query bank trained against explicit clinical concepts: the substrate for a grounded representation contract.

BioClinical ModernBERT-base · 8,192 tokens160 grounded clinical conceptsMultiplicative Concept BottleneckLarger backbones under evaluation
Layer 03
Reasoning
A concept bottleneck every prediction must flow through.

The reasoning layer activates the concepts present in a note and routes prediction signal exclusively through them, making evidence a requirement rather than an afterthought.

Per-note concept activationConcept-grounded code predictionConfidence and alternativesCSTPR telemetry
Layer 04
APIs
One integration shape, many products downstream.

Every product returns predictions, evidence, concepts, and alternatives through the same response shape—one integration surface for evolving clinical workflows.

REST endpoints and JSON responsesToken-based authenticationPer-prediction audit logsAsync webhooks on roadmap
Live today
ShifaMind
The first product built on the shared stack.
Explore ShifaMind
Build on the platform

Integrate ShifaMind today. Build on what's next.

If your clinical workflow needs grounded reasoning, partner deployments can shape the next product on the stack.