Scenario generator
Terrain, facilities from casualty collection points to Role 3 hospitals, ground and air corridors, moving air-defence, jamming and drone threats, and a casualty population. Every setting is adjustable.
For evacuation planners and exercise teams · synthetic data only
Plan casualty evacuation exercises in minutes, not weeks.
AegisMED builds realistic battlefield medical scenarios: facilities, corridors, moving threats and casualties whose condition worsens over time. Shape them on a map, keep every version, and let an AI assistant draft them from plain English.
Access is by approval. No real patient, unit or location data is ever used.
- 5
- ready-made scenario templates
- 7
- approved AI models, non-PRC origin
- 100%
- synthetic data: no real patients or units
- sha256
- every version rebuilds byte for byte
Features
Everything an exercise cell needs to build the scenario
From a one-line idea to a complete, repeatable theatre with casualties, threats and a timeline.
Edit on the map
Move, add or remove facilities, block or open corridors, and drop in casualties where the exercise needs them. Changes preview live before you save.
Timeline events
Close a road at minute 30, switch on a missile site after an hour, add a jammer in the north. Events play out across the exercise clock.
Casualty digital twin
Each synthetic casualty has vitals and a criticality score that worsens over time with bleeding and injury type, driving triage categories T1 to T4.
Versions you can trust
Every save is a new, immutable version with a fingerprint. Restore or fork any version; the same recipe always rebuilds exactly the same scenario.
AI drafting assistant
Describe an exercise in plain English and get a complete scenario to review. It also suggests escalation events and writes the exercise briefing.
The console
One screen for the whole scenario
The map shows every facility, corridor risk at the minute you choose, missile coverage, jammers and drones. The side panel holds parameters, the timeline, the AI assistant and version history.
aegismedsi.com/scenarios/…
Mountain valley exercisev4
30 facilities85 corridors170 casualtiesT1 24T2 55
MapParamsTimelineAIHistory
Map edits
1Block CCP-01 – LZ-04
2Add 20 torso-bleed casualties at CCP-02
3Move ROLE2-01 to 60.0, 20.0 km
Timeline
30–90 min · Corridor LZ-09 – ROLE1-03 closed
60 min → · SAM site (20 km) activated, north
How it works
From request to briefing in four steps
- 1
Request access
Register with your work email. An administrator approves your account and organisation.
- 2
Start a scenario
Pick a template such as mass-casualty surge or contested airspace, or describe it to the AI.
- 3
Shape it
Tune parameters, edit the map and schedule events. Watch the preview update as you go.
- 4
Save and brief
Save a version, generate the briefing, and share it with your exercise cell.
AI assistant
Describe the exercise. Review the draft. Save it.
Type something like “mountain valley, heavy drone activity, 150 casualties mostly torso injuries, eastern road closed after 30 minutes”. The assistant turns it into a complete scenario for you to check.
› Describe the scenario
Mountain valley, heavy drone activity, 150 casualties mostly blast torso injuries, a key road in the east closed after 30 minutes for an hour, 3 hour exercise.
› Draft
- terrain: mountainous · drones: high
- casualties: 150 · torso 70% · head 10% · limb 20%
- duration: 180 min
- event: corridor closed 30–90 min, east
- Choices are bounded: the model selects terrain, threat levels, counts and events; the engine supplies every number.
- Suggests escalation events and writes a plain-text exercise briefing.
- Runs on Ollama Cloud or on your own hardware; organisations choose the model.
- Every request is rate limited per organisation and recorded for audit.
Approved cloud models
gpt-oss
OpenAI · US
Gemma 4
Google · US
Mistral Large 3
Mistral AI · FR
Nemotron 3
NVIDIA · US
Only models whose base weights come from developers outside the PRC. Enforced by the server, not just the interface.
Security
Built for organisations that take access seriously
Controlled access, separated data and a record of who did what.
Approval-only access
No self-service accounts and no default logins. Every person is approved by an administrator before they can sign in.
Isolated organisations
Each organisation's data is separated inside the database itself (row-level security), not just in the application.
Tamper-resistant audit log
Sign-ins, approvals, role changes and AI settings are recorded in an append-only log the application cannot rewrite.
Protected keys
Organisation AI keys are encrypted at rest. Sessions use secure, HTTP-only cookies; sign-in attempts are rate limited.
Your choice of AI hosting
Use Ollama Cloud, or keep every prompt on your own hardware with a self-hosted model.
Operational visibility
Structured logs, health checks, metrics and an admin overview of usage, errors and pending approvals.
Roles
The right view for each member of the team
Command staff
Design and edit scenarios, run the AI assistant, save versions.
Triage officer
Review casualty populations, triage mix and the exercise timeline.
Flight medic
Study the theatre, corridors and threats before the exercise.
System admin
Manage the organisation's users, AI model choice and activity log.
Roadmap
Coming next: risk-aware evacuation routing
These capabilities are in development and not yet available.
Corridor risk model
A graph neural network scores each corridor's threat from the live threat picture.
Trade-off routes
An optimiser returns the set of best routes balancing casualty deterioration against threat exposure, so a person makes the final call.
Live casualty telemetry
Streaming vitals and time-to-critical countdowns for each casualty during the exercise.
FAQ
Questions teams ask first
Is any of the data real?
No. Every scenario, facility, casualty and threat is machine-generated for training. Never enter real patient, unit or location data.
Can it be used for real triage or operations?
No. The physiological model uses engineering values chosen for simulation and is not clinically validated. AegisMED is a planning and training tool.
Which AI models can we use?
Only models whose base weights come from developers outside the People's Republic of China: OpenAI gpt-oss, Google Gemma, Mistral, NVIDIA Nemotron, and for self-hosting Meta Llama, Microsoft Phi and IBM Granite. The server enforces this list.
Can the AI invent numbers?
No. The assistant only picks from bounded choices (terrain, threat levels, casualty counts, events). All physiology, risk and coordinates come from the deterministic engine, and you review every draft before saving.
How do we get access?
Request access with your work email. A platform administrator approves the request and assigns your organisation and role.
Ready to build your next exercise?
Request access with your work email. An administrator will review it.