diff --git a/README.md b/README.md
index 4a3b5e3db..407b3a87a 100644
--- a/README.md
+++ b/README.md
@@ -20,6 +20,81 @@ systems.
 
 CBRAIN (and the alternative NeuroHub interface) consists of two Ruby on Rails Applications: BrainPortal and Bourreau
 
+## System architecture
+
+```mermaid
+%%{init: {"theme": "base", "flowchart": {"defaultRenderer": "elk"}}}%%
+flowchart LR
+  Users([Researchers & Web browsers])
+  NH[NeuroHub Portal<br/>Alternative UI]
+  subgraph Frontend[Frontend resources]
+    BP[BrainPortal<br/>Rails frontend]
+    DB[(Shared database & metadata)]
+  end
+  subgraph ExternalResources[External resources]
+    direction TB
+    subgraph DataProviders[Data providers]
+      direction TB
+      DP1[Data provider<br/>S3/HTTP/FTP, etc.]
+      DP2[Data provider<br/>S3/HTTP/FTP, etc.]
+      DP3[More...]
+    end
+    subgraph HPCResources[HPC resources]
+      direction TB
+      subgraph ResourceA[HPC resource A]
+        BO1[Bourreau]
+        Sched1[HPC scheduler<br/>SLURM/PBS/...]
+        subgraph ComputePoolA[Compute nodes]
+          Compute1[Compute node]
+          Compute2[Compute node]
+          ComputeMoreA[More...]
+        end
+        Scratch1[(Working directories<br/>Shared storage)]
+      end
+      subgraph ResourceB[HPC resource B]
+        BO2[Bourreau]
+        Sched2[HPC scheduler<br/>SLURM/PBS/...]
+        subgraph ComputePoolB[Compute nodes]
+          Compute3[Compute node]
+          Compute4[Compute node]
+          ComputeMoreB[More...]
+        end
+        Scratch2[(Working directories<br/>Shared storage)]
+      end
+    end
+  end
+
+  Users --> BP
+  Users --> NH
+  NH --> BP
+  BP --> DB
+  BP --> DataProviders
+  BP -->|SSH/XML| BO1
+  BP -->|SSH/XML| BO2
+  BO1 --> DataProviders
+  BO2 --> DataProviders
+  BO1 --> DB
+  BO2 --> DB
+  BO1 --> Sched1 --> ComputePoolA
+  BO2 --> Sched2 --> ComputePoolB
+  BO1 --> Scratch1
+  BO2 --> Scratch2
+  ComputePoolA --> Scratch1
+  ComputePoolB --> Scratch2
+```
+
+At a high level, researchers interact with BrainPortal (or the NeuroHub
+portal) through a web browser. The arrows in the diagram show the primary
+request flow; responses are implied by each call. BrainPortal orchestrates
+access to data providers, persists metadata in the shared database, and
+delegates execution requests to one or more Bourreau instances, typically
+over SSH/XML. Each Bourreau runs on a specific HPC resource and connects
+to the local scheduler to launch jobs on that resource's compute
+nodes. Bourreaux manage working directories on shared storage, synchronize
+job and file state back to the database for BrainPortal to display, and
+fetch or stage data from multiple data providers as part of backend task
+execution.
+
 ## BrainPortal
 
 BrainPortal is the frontend of the CBRAIN architecture. It is a
@@ -45,4 +120,3 @@ running on the HPC where it resides.
 ## For more information
 
 CBRAIN is extensively documented in its [Wiki](https://github.com/aces/cbrain/wiki).
-