live walk-through scene 01 / 06
01 The analyst clicks the chat surface and starts typing a question.
YOUR TENANT · YOUR VPC · YOUR ENCRYPTION KEYS SOC 2 · ISO 27001 · GDPR · GxP USERS Domain analyst asks regulator-grade questions Your copilot embeds the agent API Compliance officer audits, exports the trace Executive reads the dashboard, signs off Regulator arrives with questions LAYER 05 · CONSUMPTION SURFACES · WE DELIVER Same graph, four ways. Humans ask. Agents call. Compliance audits. Executives report. Citable answer UI · API · MCP server Conversational analytics drill from chart to entity Agent grounding API your copilots embed it Audit and lineage explorer regulator-grade trace · exportable LAYER 04 · VERTICAL ONTOLOGY PACK · WE DELIVER The whole foundation, pre-configured for your industry. Six weeks to production. Pharma reg affairs LIVE Water utilities LIVE Building safety LIVE Insurance claims Q3 2026 + your vertical LAYER 03 · SEMANTIC LAYER · ONTOLOGY + CONTEXT GRAPH Binds catalogue metadata to typed entities. Builds the context graph your AI can reason against. ONTOLOGY BINDING raw fields to typed entities CONTEXT GRAPH queryable · walkable · citable PROVENANCE VERSIONED historical answers reproducible RBAC ON GRAPH predicate-level access control REGULATOR-READY LAYER 02 · MANAGED CATALOGUE · WE DELIVER [branded as yours] A complete catalogue in your tenant. Crawls, classifies, profiles, indexes. Always on. Schema scan column profile · lineage PII classification Glossary mapping stewards · domains freshness check Metadata index 12.4M assets · 847 dashboards 2341 tags · 186 owners Quality & trust rule library drift detection Access control policy · audit key vault LAYER 01 · YOUR DATA SOURCES · YOUR ENVIRONMENT Snowflake Databricks Veeva Vault Argus Safety Guidewire Maximo SAP + 70 source connectors ANSWER · GROUNDED · 5 SOURCES · 4 GRAPH PATHS Two Phase 3 oncology trials show signals within the FDA's 2025 immunogenicity guidance: CT-2024-118 and CT-2024-202. Both need updated IND amendments.
cursor and particles show how a question becomes a citable answer
Cursor
The analyst's actions: clicking surfaces, typing questions.
Query down
The question travels from the chat surface down through the pack and into the graph.
Data up
Source content flows into the catalogue, then binds into the typed graph.
Citation
The answer returns with the reasoning trace and source provenance.
Substrate Layer 02 is powered by open-source DataHub, the catalogue trusted by LinkedIn, Pinterest and Coinbase. We deploy and manage it inside your tenant, branded as part of your foundation. Honest about the substrate, opinionated about everything we build on top.
/ WHY THESE LAYERS

The order matters. Catalogue first, graph second, agents last.

The catalogue gives the agent a map.

Without one, your model is guessing which Snowflake table has the customer record. A managed catalogue, crawled and classified every night, is the floor every other layer stands on.

The ontology gives the catalogue meaning.

A field called “cust_id” means nothing to a model. Bound to a typed customer entity, with relationships to claims, policies and reserves, it becomes something the agent can actually reason against.

The graph gives the model a path to walk.

Multi-hop questions ("which claims sit below benchmark and trigger reinsurance") need traversal, not retrieval. A vector store cannot do this. A typed, queryable graph can, and every step is auditable.

The agent layer gives the regulator an artefact.

Every answer comes back with the reasoning trace, the citations and the audit log. The model becomes a tool that produces evidence, not just text. Your QA reviewer can audit. Your compliance officer can export.

See the architecture in your environment. Connected to your data.

Thirty minutes. No deck, no demo theatre. Bring your hardest stuck pilot.