Beyond Dashboards.Beyond Chatbots.
Zynomi's Agentic Analytics is designed as a crew of bounded agents — not a single chatbot. A supervisor that owns no data tools routes each question to a specialist bound to exactly one engine: the governed semantic layer or the temporal knowledge graph (clinical ontology + GraphRAG), connected through MCP. Every number is traceable back to source data.
Watch the AI Agent in Action
What is Agentic Analytics?
The Next Evolution in Data Intelligence
The term 'AI' is everywhere, but most tools hit a ceiling: one chatbot, one prompt, one answer. Real data work isn't about single questions — it's about multi-step investigation and reasoning. Zynomi's Agentic Analytics is designed as a crew of bounded specialists routed by a supervisor, each grounded in exactly one engine — so answers reason and act with rich context while staying traceable. Digital teammates, not just tools.
Reason
Deep Context Understanding
A simple chatbot might translate 'show me enrollment' into SQL. But what does 'enrolled' mean? Is it consented? Screened? Active? An agent grounded in the semantic layer knows precisely how metrics are calculated — and an agent grounded in the temporal knowledge graph knows how entities connect and how they evolved over time.
Act
Multi-Step Analysis
The most significant leap from a chatbot to an agent is the ability to act on a goal through a series of steps. Ask for 'month-over-month enrollment growth' and the agent plans, derives the metric, and executes — even if that specific metric doesn't exist yet.
Expand
Growing Semantic Context
As your needs evolve, simply add new Cube models to expand the semantic layer. The LLM automatically gains access to richer metadata — new dimensions, measures, and joins — enabling increasingly sophisticated queries without any ML training or fine-tuning.
A Crew of Bounded Agents
One Supervisor. Four Specialists. Exactly One Engine Each.
Zynomi's agent fleet is designed so the supervisor owns no data tools — it classifies your question and routes it to a bounded specialist. Each specialist sees only its own engine's tools, everything is read-only by construction (one gate, tested in CI, not promised in a prompt), and tools compute while the model narrates.
Supervisor
Owns no data tools — classifies the question and routes it to bounded specialists
Trial Metrics Analyst
Governed KPIs only
Semantic Layer
Centrally defined metrics
Study Historian
Fixed-shape Cypher via MCP
Temporal Graph
Relationships + time
Operations Agent
Read-only lookups
Transactional Store
Current operational state
Data Steward
Quality & lineage checks
Quality / Lineage
Tests and lineage metadata
One engine per specialist
Each agent sees only its own context's tools
Read-only by construction
One gate, tested in CI — not promised in a prompt
Tools compute, the model narrates
Agents never re-derive a number, never invent an edge
Three classes of questions, one front door
“What was the enrollment rate across Phase III studies last quarter?”
Routed to the Trial Metrics Analyst — computed from governed KPIs in the semantic layer.
“Show everything that happened after Amendment 4 — who requested it, who approved it, which sites were affected.”
Routed to the Study Historian — one traversal of the temporal knowledge graph.
“Which sites affected by Amendment 4 saw more protocol deviations afterward?”
The graph finds the affected sites, the analytical layer computes before/after deviation rates, and the LLM composes one explanation.
The Complete Agentic Stack
From Raw Data to Intelligent Insights
Zynomi CTMS ships a ready-to-use Modern Data Lakehouse with a governed Semantic Layer, supporting Bring Your Own Data Visualizations and an Agentic Analytics AI Chatbot whose answers are generated from governed metric definitions. The entire lakehouse-to-semantic pipeline is packaged using dbt Core and Cube Core, but can be swapped to their cloud versions (dbt Cloud, Cube Cloud) for high-scale enterprise deployments.
Agent Crew & Chat Interface
React MicrofrontendA custom-built React chat client embedded as a microfrontend plugin directly in the application. Behind it, a supervisor agent routes each question to bounded specialists — a seamless, branded experience within your clinical trial management workflow.
MCP Server Bridge
Node.js + PythonModel Context Protocol connecting AI to your data
🔗 ctms-mcp-server.zynomi.com
Semantic Layer
Cube.devGoverned metrics, pre-joined views, consistent definitions
Temporal Knowledge Graph
Graphiti + FalkorDBClinical ontology + GraphRAG: entities and relationships with valid-from/valid-to on every edge — amendment provenance, affected sites, and before/after history in one traversal. In development as part of the Zynomi Agent Fabric (ZAF).
Data Lakehouse
dbt + PostgreSQLBronze/Silver/Gold medallion architecture with CDISC-compliant models. PostgreSQL is the default warehouse, with support for Snowflake, Databricks, and Redshift.
Powered by Open Standards • Built for Enterprise Scale
Key Capabilities
Everything You Need for Intelligent Clinical Analytics
Pre-joined, business-friendly views that abstract the complexity of the underlying star schema. Consistent metric definitions across all tools — from AI chatbots to Tableau dashboards.
- Pre-aggregated clinical KPIs
- CDISC-aligned data models
- Single source of truth for metrics
Semantic Layer
Pre-joined, business-friendly views that abstract the complexity of the underlying star schema. Consistent metric definitions across all tools — from AI chatbots to Tableau dashboards.
MCP Server
Model Context Protocol bridge connecting AI assistants to your clinical data. 6 built-in tools for querying, exploring, and analyzing trial data programmatically.
Grounded, Traceable AI
Answers are grounded twice: governed numbers come from the semantic layer, computed exactly as centrally defined; relationships and history come from the temporal knowledge graph (clinical ontology + GraphRAG). Tools compute, the model narrates — and every number is traceable back to source data.
Bring Your Own BI
Connect Tableau, Power BI, Metabase, or Looker to the semantic layer over its Postgres-compatible SQL API. Your analysts keep their favorite tools while benefiting from governed, consistent metrics.
Multi-Step Analysis
AI executes complex analytical workflows, not just single queries. Ask for derived metrics, comparisons across time periods, or cohort breakdowns — the agent plans and executes.
Extensible Semantic Models
Expand your analytical capabilities by adding new Cube models. The LLM automatically discovers new dimensions, measures, and relationships — delivering richer context without any ML training.
Data Lineage & Governance
From Source to Insight — Complete Transparency
Every metric, every dashboard, every AI response is traceable back to its source. Our medallion architecture (Bronze → Silver → Gold) ensures data quality at every step, while dbt provides complete lineage documentation.
Key Features
Medallion Architecture
Ready to Transform Your Clinical Analytics?
Experience AI that works from your governed clinical data definitions — answers grounded in the semantic layer and traceable to source.