A2A architecture proposal #2
nikomoegre
started this conversation in
Ideas
Replies: 1 comment
-
graph TD
%% User at the top
User[👩⚕️ Clinical User]
%% Main Orchestrator
subgraph Orch["🎯 ORCHESTRATOR AGENT"]
OrchState["Mental State:<br/>• Memory (M_t^mem)<br/>• World Model (M_t^wm)<br/>• Goals (M_t^goal) - derived from user input<br/>• Reward (M_t^rew) - RL with GRPO"]
end
%% Primary agents
subgraph Sem["🧠 SEMANTIC AGENT"]
SemState["Mental State:<br/>• Memory (M_t^mem)<br/>• Graph World Model (M_t^wm) - from medical<br/> vocabularies enriched with context<br/>• Goals (M_t^goal) - derived from incoming queries<br/>• Reward - Graph of Thoughts"]
end
subgraph SQL["💾 SQL AGENT"]
SQLState["Mental State:<br/>• Memory (M_t^mem)<br/>• OMOP CDM Knowledge (M_t^wm)<br/>• Goals (M_t^goal) - derived from incoming queries<br/>• Reward - RL"]
end
%% Databases at the bottom
EHR[(🏥 EHR System<br/>OMOP CDM)]
Vocab[(📚 Medical Vocabularies<br/>OHDSI OMOP)]
%% Shared functions box (on the left side)
subgraph Core["SHARED AGENT FUNCTIONS"]
Functions["All agents implement:<br/>• Perception (P)<br/>• Cognition (C)<br/>• Action Execution (E)<br/>• Learning (L)<br/>• Reasoning (R)"]
end
%% Other agents box (on the right side, disconnected)
subgraph Others["OTHER AGENTS"]
OtherList["Additional specialized agents:<br/>• ⚖️ Ethics Agent<br/>• 🩺 Clinical Decision Agent<br/>• 🔬 Research Agent"]
end
%% Main flow connections
User -->|Query| Orch
Orch -->|Response| User
%% Orchestrator to main agents
Orch -->|Delegates| Sem
Orch -->|Delegates| SQL
SQL -->|Returns clinical insights| Orch
%% Agent to database connections
SQL -->|Queries| EHR
Sem -->|Lookups| Vocab
%% Inter-agent connections
Sem -->|Provides medical concepts| SQL
SQL -->|Queries for semantic context| Sem
%% Styling - simple two-color scheme
classDef agent fill:#e3f2fd,stroke:#1565c0,stroke-width:2px
classDef database fill:#f5f5f5,stroke:#757575,stroke-width:2px
classDef user fill:#ffffff,stroke:#000000,stroke-width:2px
classDef info fill:#fffde7,stroke:#f57f17,stroke-width:1px,stroke-dasharray: 5 5
class Orch,Sem,SQL agent
class EHR,Vocab database
class User user
class Core,Others info
```
|
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Apologies in advance for the - a bit lengthy - one-pager, heavily inspired by the lengthy https://arxiv.org/pdf/2504.01990. I wanted to give an overview on current notions and definitions in the agent space and adapt them to our framework. Further I wanted to give a first overview on how our framework could look like and what the actual advantages of incorporating A2A besides MCP could be.
Argument on agentic frameworks in live EHR querying.docx
Beta Was this translation helpful? Give feedback.
All reactions