AI INTERN @ ONIX NETWORKINGHEALTHCARE

Agentic Systems for Synthetic
Healthcare Data
Generation

Validating Agentic Workflows as a superior alternative to statistical methods. Orchestrating autonomous agents to generate FHIR-compliant patient records.

01. Executive Summary

The objective of this initiative was to validate agentic systems as a superior alternative to traditional statistical methods for synthetic data generation. Healthcare was selected as the pilot domain due to its high demand for data privacy and the structured nature of patient-practitioner interactions.

The Strategic Goal

Deliver a system capable of simulating complex interactions to produce realistic, standard-compliant (FHIR) data, serving as a stepping stone for the organization's pivot toward agentic workflows.

FHIR Compliant

Validated via fhir.resources

10k+ Records

Enriched Harvard Dataset

LangGraph Core

LLM-Driven Orchestration

02. Technical Architecture

The final Agentic System moved beyond the probabilistic limitations of our earlier Mesa prototypes. The quality of the simulation depended on a "Contextual Backbone" feeding into an intelligent orchestrator.

Knowledge

Enriched Harvard KB

10,000+ Records defining the "World State".

Orchestrator

LangGraph Agents

LLM-driven decision making. Handles graph state memory and tool execution.

Output

Standardized Data

FHIR-compliant JSON resources ready for ingestion.

03. Execution Roadmap

Phase 1: Validation (Rule-Based)

Established the Knowledge Base utility...

Phase 2: The Baseline (Mesa Simulation)

Developed a separate agent-based model using Mesa...

Phase 3: Intelligent Orchestration

The final evolution. Replaced statistical rules with LLM-driven Agents...

04. Outcome & Impact

Technical Success

Successfully validated that Agentic Workflows...

Ecosystem Fit

Demonstrated how agent-based modeling fits...

Ready to see the code?

The architecture is modular and ready for inspection.