01. Executive Summary
As a leader in enterprise data migration, Onix sought to future-proof Kingfisher, their flagship synthetic data platform. The vision was to transition Kingfisher from relying on static sample data and schemas into a fully prompt-driven agentic system (e.g., "Generate a synthetic pipeline for a pediatric hospital").
Healthcare was selected as the strategic pilot domain. While healthcare interactions are complex and highly regulated, patient journeys follow deterministic and predictable paths, making it the ideal proving ground to validate AI decision-making logic.
The Strategic Goal
Align technical execution with executive vision by proving that agentic workflows could generate hyper-realistic, FHIR-compliant data entirely on demand. This pilot was the necessary catalyst to secure engineering buy-in and transition Kingfisher into a holistic, AI-first product.
FHIR Compliant
Validated via fhir.resources
20+ Entities
Deterministic patient branching
LangGraph Core
LLM-Driven Orchestration
02. Technical Architecture
Realism hinged on the quality of the Knowledge Base (KB). Instead of a conventional orchestrator, I designed a dynamic architecture powered by LangGraph, drawing from a semantically enriched backbone to simulate logical healthcare outcomes.
Enriched Harvard Dataset
10,000-record FHIR dataset enriched with derived relationships and attributes across 20+ distinct entities.
LangGraph Agents
Invokes specialized LLM-driven agents and tools. Graph state handles shared memory for deterministic branching.
Standardized Data
FHIR-compliant records simulating end-to-end patient and insurance workflows on demand.
03. Execution Roadmap
Phase 1: Rule-Based Validation
Validated the scope of the KB by creating deterministic, rule-based interactions to confirm foundational logic.
Phase 2: Mesa-Based Simulations
Moved to a dynamic environment using the Mesa framework. Simulated structured interactions between agents with predefined goals (but no embedded intelligence).
Phase 3a: LangGraph Agentic System
Combined prior learnings with an LLM-driven decision-making layer. User requests were interpreted, planned, and executed automatically through specialized agent orchestration.
Phase 3b: Insurance Domain Integration
Expanded the simulation logic to cover complex insurance models, including claims, adjudication, and benefits processing, demonstrating absolute domain adaptability.
04. Outcome & Impact
Sales Enablement
Directly supported the go-to-market strategy for Kingfisher's new capabilities. Powered live executive demonstrations that anchored engagements with 6 distinct enterprise healthcare and insurance clients.
Strategic Pivot Validated
Successfully validated the agentic thesis for Onix leadership, proving Kingfisher could securely pivot from relying on static client schemas to an on-demand, prompt-driven product.