A risk-intelligence platform exploring how complex systems can be modelled before decisions are made.
Explore the platform · Architecture · Modules · API · Security · Roadmap
Pick a starting point — each panel expands right here in the README.
🌐 Platform overview
Prexus is an early-stage, multi-module intelligence platform focused on risk modelling, scenario analysis, and probabilistic decision support across climate, infrastructure, and other complex domains.
Design principle: make uncertainty explicit; do not present forecasts as certainty.
🧮 How the simulation flow works
flowchart TD
A["Define assets & assumptions"] --> B["Ingest available signals"]
B --> C["Build scenario distributions"]
C --> D["Run simulation engine"]
D --> E["Summarise risk metrics"]
E --> F["Review evidence & limitations"]
F --> G["Decision support"]
style A fill:#102A43,stroke:#55D6FF,color:#EAFBFF
style D fill:#102A43,stroke:#55D6FF,color:#EAFBFF
style G fill:#102A43,stroke:#55D6FF,color:#EAFBFF
The diagram is a conceptual flow, not a claim that every source or model is enabled in every deployment.
🧭 Jump into a module
| Module | Focus | Current framing |
|---|---|---|
| Meteorium | Physical climate exposure and risk metrics | First climate-risk module |
| Raksha | Threat-signal correlation and situational awareness | In development |
| Healtho | Population-health and system-load scenarios | Planned / in build |
🔌 API quick peek
| Need | Endpoint |
|---|---|
| Check service liveness | GET /health |
| Check readiness | GET /ready |
| Create an account | POST /register |
| Sign in | POST /login |
| Submit an access application | POST /apply |
| Read applications (admin permission required) | GET /admin/applications |
See the full API reference. Actual availability depends on the deployed gateway configuration.
🛡️ Security & maturity notes
- Authentication and role-based permissions are implemented in the API gateway.
- The repository includes automated security and dependency checks.
- Some capabilities remain planned or in development; see the capabilities matrix.
- This README is not a security certification, production-readiness claim, or guarantee of forecast accuracy.
Prexus is an AI-driven system designed to predict large-scale risks and outcomes across complex systems such as infrastructure, geopolitics, and institutional decision-making.
Instead of reacting to events, Prexus focuses on anticipating them.
Modern systems are becoming:
- Highly interconnected
- Increasingly unpredictable
- Difficult to manage using traditional models
Governments and institutions today rely on reactive strategies.
Prexus aims to shift this from reaction → prediction.
- Early-stage prototype
- Monte Carlo simulation engine (Python + Rust)
- Scenario-based risk modeling
- Probabilistic outcome forecasting
Input:
- System variables (economic, infrastructure, external risks)
Output:
- Probability of specific events
- Simulation paths across multiple scenarios
- Risk distribution over time
1. Define system variables
2. Run thousands of simulations
3. Analyze probability distributions
4. Generate predictive insights
| Component | Status |
|---|---|
| Core simulation engine | ✅ Functional |
| Prototype | ✅ Completed |
| Meteorium (Climate Risk) | 🧪 Prototype / deployment status must be verified |
| Real-world dataset integration | 🔄 Expanding |
| Healtho (Health Intelligence) | 📝 Planned; no implementation claimed |
| Raksha (Threat Intelligence) | 🧪 Prototype implemented; uncalibrated |
To build a sovereign intelligence layer that enables:
- Governments to predict risks before they occur
- Institutions to make high-stakes decisions with data-backed foresight
- Systems to evolve from reactive → predictive
- Python + FastAPI
- Rust (Monte Carlo simulation core)
- Go (API Gateway)
- Simulation modeling
- Probabilistic analysis
- Improve model accuracy
- Integrate real-world datasets
- Build scalable architecture
🧱 Expand the architecture overview
Prexus is built as a distributed, polyglot system — each layer uses the best-fit language for its role.
┌─────────────────────────────────────────────────────────────┐
│ CLIENT LAYER │
│ Gov Dashboards · Financial Terminals · Enterprise │
└──────────────────────────┬──────────────────────────────────┘
│ HTTPS / TLS 1.3
┌──────────────────────────▼──────────────────────────────────┐
│ API GATEWAY (Go) │
│ JWT Auth · ABAC · Rate Limiting · CORS · Audit Log │
└───────────────┬─────────────────────────┬───────────────────┘
│ │
┌───────────────▼──────────┐ ┌───────────▼───────────────────┐
│ INTELLIGENCE LAYER │ │ COMPUTE LAYER │
│ (Python) │ │ (Rust) │
│ · Risk Analytics │◄─► · Monte Carlo Engine │
│ · Scenario Models │ │ · VaR / CVaR │
│ · IPCC Pathways │ │ · Numerical Analysis │
└───────────────┬──────────┘ └───────────────────────────────┘
│
┌───────────────▼──────────────────────────────────────────────┐
│ DATA ADAPTER LAYER │
│ Sentinel-1 SAR · ECMWF · Bloomberg · IPCC AR6 │
└───────────────────────────────────────────────────────────────┘
| Layer | Language | Role | Key Capability |
|---|---|---|---|
| API Gateway | Go | Request routing, auth, audit | Zero-trust ABAC, JWT, rate limiting, SHA-256 hash chain |
| Intelligence Engine | Python / FastAPI | Analytics orchestration | Risk scoring, scenario modelling, IPCC pathway integration |
| Compute Acceleration | Rust | High-performance numerics | Monte Carlo simulation, VaR/CVaR, loss distribution |
| Data Adapters | Python | External data ingestion | Climate, environmental, financial signal normalisation |
| Audit Ledger | Go | Immutable event log | SHA-256 hash-chained tamper-evident records |
Prexus is a multi-module intelligence platform. Each module addresses a distinct institutional risk domain. They share a common compute layer, auth infrastructure, and audit ledger.
Status: Live
Meteorium is the environmental intelligence core of Prexus — a dedicated risk computation engine for physical climate exposure analysis. It is the first module in production.
┌─────────────────────────────────────────────────────┐
│ METEORIUM ENGINE │
├──────────────────────────┬──────────────────────────┤
│ Input Parameters │ Intelligence Output │
├──────────────────────────┼──────────────────────────┤
│ · Asset coordinates │ · Composite Risk Score │
│ · Asset valuation │ · VaR 95% │
│ · Prediction horizon │ · CVaR 95% │
│ (180d / 1y / 3y) │ · Expected Loss │
│ · Urban density factor │ · Risk Band │
│ · Climate scenario │ · Audit Receipt │
│ · Insurance coverage │ │
│ · Liquidity shock factor │ │
└──────────────────────────┴──────────────────────────┘
STOCHASTIC SIMULATION CORE
┌───────────────────────────────────────────┐
│ 10,000 Monte Carlo iterations │
│ IPCC AR6 scenario integration │
│ Urban density amplification (λ) │
│ Insurance drag factor (δ) │
│ Liquidity shock multiplier (κ) │
└───────────────────────────────────────────┘
Supported Climate Scenarios
| Scenario | ID | Description | Risk Premium |
|---|---|---|---|
| 🟢 Baseline | baseline |
Orderly, Paris-aligned policy | +0% |
| 🟡 Disorderly Transition | disorderly |
Delayed policy action, repricing shock | +9% |
| 🔴 Failed Transition | failed |
No policy correction, full physical exposure | +16% |
Prediction Horizons
| Tactical | Strategic | Structural |
|---|---|---|
| 180 Days | 1 Year | 3 Years |
| Near-term positioning | Capital planning | Long-run mispricing |
Risk Band Classification
| Score | Band | Indicator |
|---|---|---|
| ≥ 0.85 | CRITICAL |
Immediate exposure — intervention required |
| ≥ 0.75 | HIGH |
Elevated repricing risk — review urgently |
| ≥ 0.60 | ELEVATED |
Material risk — monitor closely |
| ≥ 0.50 | MODERATE |
Acceptable range — standard monitoring |
Platform intelligence view. Add asset to see risk visualisation.
Screenshot / demo recording coming soon. The globe renders live climate risk heatmaps, asset pins with severity-graded warning tabs, RCP 8.5 scenario projection (2023–2050), and Meto AI — a 3-model intelligence assistant (Claude / GPT-4o / Gemini) with full portfolio context.
Status: In Build
Healtho applies the Prexus simulation core to population health and bio-systemic risk domains. Designed for national health authorities, pandemic preparedness agencies, and insurance actuaries.
Planned capabilities:
- Epidemic spread modelling across urban networks
- Healthcare system load forecasting under stress scenarios
- Mortality and morbidity risk curves (Monte Carlo)
- Bio-systemic shock propagation across economic sectors
- Integration with WHO datasets and national health registries
Target clearance: Level 3 · Target deployment: National governments, Central health authorities
Status: Prototype implemented · 31 tests documented
Raksha is being built to correlate cyber, geopolitical, and physical-hazard signals into explainable risk assessments for assets, organisations, and infrastructure. The current prototype is more than a roadmap, but it is not yet a calibrated operational intelligence service.
Implemented in the repository:
- STIX 2.1 data model with Prexus event and asset objects, deterministic identifiers, Traffic Light Protocol (TLP) markings, source provenance, and versioned records.
- Offline-first ingestion with synthetic fixtures and bundle building/validation.
- Connector parsers for CISA Known Exploited Vulnerabilities (KEV), FIRST Exploit Prediction Scoring System (EPSS), GDELT 2.0 event data, and NASA FIRMS VIIRS fire detections.
- Read-only TAXII 2.1 interface for discovery, collections, object retrieval, filtering, and pagination.
- Attribute-based access control (ABAC) using default-deny and deny-overrides, clearance, compartments, TLP, purpose limitation, export rules, and no-write-down controls.
- Transparent v0 scoring with evidence references and uncertainty intervals; scores explicitly remain
calibrated:false. - Dependency cascade simulation and a rolling-origin back-test harness for Brier score, expected calibration error (ECE), reliability bins, and comparison with climatology.
Important limitations: the documented test suite uses synthetic fixtures; no live-feed fetch was verified in the development environment. The GDELT column mapping still needs validation against a live sample. Scores are uncalibrated and must not be presented as validated forecasts. OpenCTI/MISP interoperability, Postgres/PostGIS storage, analyst UI, and real-history back-testing remain open work.
See Raksha design and implementation notes and the audit record.
All routes are served by the Go gateway; the Python engine is a private service reachable only from the gateway.
Authentication: Authorization: Bearer <JWT> (24 h). Rate limit: 5 req/s per IP (burst 10); /apply: 3 per 2 min.
| Method · Path | Auth | Purpose |
|---|---|---|
GET /health · GET /ready |
none | liveness · readiness (database) |
POST /register · POST /login |
none | create account (role user) · obtain JWT |
POST /apply |
none | access / procurement application (stored in Postgres, emailed if SMTP configured) |
GET /admin/applications[/:ref] |
admin (applications:read) |
read stored applications with bounded pagination and reference validation |
GET/POST/PUT/DELETE /assets[/:id] |
assets:* |
the caller's assets |
GET /alerts |
assets:read |
alerts derived from the caller's asset scores |
POST /risk/asset · /risk/portfolio · /risk/stress-test |
risk:run |
Monte Carlo (Rust core) via the engine |
POST /risk/simulate |
risk:run |
scenario simulation (catmodel-v0, uncalibrated) |
GET /risk/health · /sources · /lake/stats · /lake/files |
risk:run |
engine status and data catalogue |
POST /claude · /openai · /gemini · /chat · /analyze |
risk:run + daily quota |
the single LLM gateway (model allowlist, *_BASE_URL for local models) |
GET/PUT /me · GET /conduit/tools |
auth | profile · Conduit MCP tool list |
Errors: 400 validation · 401 missing/invalid token · 403 permission · 404 · 413 payload too large ·
429 rate/quota · 502/503 upstream unavailable (details are logged, never returned).
| Layer | Implemented |
|---|---|
| Transport / CORS | explicit origin allow-list, no credentials, TLS at the platform edge |
| Authentication | JWT (HS256, 24 h, algorithm pinned); bcrypt password hashing (cost 10) with a dummy-hash timing defence |
| Authorisation | RBAC (admin/user/viewer → permissions). ABAC (clearance, compartments, TLP, purpose) lives in packages/prexus_core and is enforced by Raksha |
| Abuse control | per-IP rate limit (5 req/s, burst 10), strict /apply limiter, per-user daily AI quota, body-size caps |
| Service-to-service | engine is a private service and requires ENGINE_SECRET on every route; it refuses to boot without it |
| Audit | tamper-evident ledger (SHA-256 chain + Ed25519 signed head) in air-gap, Raksha, and the Go gateway (production requires durable ledger storage and a private signing key) |
| Supply chain | CodeQL (4 languages), govulncheck, pip-audit, cargo audit, gitleaks, SLSA provenance for release artifacts |
Not yet implemented: token revocation / refresh tokens, per-organisation tenancy in the gateway, HSM-backed key management,
third-party penetration test. See docs/AUDIT_FIXES.md.
Prexus includes an air-gap control-plane prototype designed for operator workflows in restricted-network environments. Its goal is to move approved, signed data bundles into a controlled enclave without requiring the enclave to fetch external feeds directly.
Implemented and documented:
- Per-operator authentication for the control-plane console.
- Signed, hash-chained audit records with an Ed25519-signed ledger head and an offline verification path.
- Signed-bundle import flow with signature, rollback, and file-hash checks before parsing and storage.
- An engine node restricted to internal addresses, plus an egress self-test.
- Automated control-plane tests, including a run of the same suite in a network namespace with only loopback access.
How the intended flow works
flowchart LR
A["External feeds / source adapters"] --> B["Validate & build bundle"]
B --> C["Sign bundle"]
C --> D["Controlled transfer"]
D --> E["Verify signature, rollback & hashes"]
E --> F["Enclave store + audit ledger"]
F --> G["Authorised queries"]
style C fill:#102A43,stroke:#55D6FF,color:#EAFBFF
style E fill:#102A43,stroke:#55D6FF,color:#EAFBFF
style F fill:#102A43,stroke:#55D6FF,color:#EAFBFF
This is a tested development implementation, not an air-gap certification or a claim that every deployment is physically disconnected. The runbook still needs deployment-specific verification. SSO/PKI integration, high availability, and hardware security module (HSM)-backed key management remain open. The Go gateway now has signed file-ledger middleware, but its ledger is not transactionally coupled to PostgreSQL mutations and requires deployment-specific durable storage and key provisioning. See air-gap overview, runbook, gateway ledger runbook, ledger specification, and audit/fix status.
The repository's audit record documents these passing test totals in its authoring sandbox:
| Area | Recorded tests | Scope / caveat |
|---|---|---|
| Go API gateway | 32 unit + 5 PostgreSQL integration | Includes race-clean unit run and real PostgreSQL 16 integration |
| Python engine | 48 | Engine tests |
prexus_core |
30 | Includes shared ABAC primitives |
| Raksha | 31 | Synthetic fixtures; does not prove live-feed quality or calibration |
| Node air-gap control plane | 30 | 28 standard + 2 run in a no-network namespace |
| Rust | 23 | Simulation-core tests |
These are the counts recorded in docs/AUDIT_FIXES.md, not a substitute for checking the current commit's CI. Dependency scanners, Docker, and deployment checks are environment-dependent; see the audit record for what was and was not run.
| Service | Platform | Role |
|---|---|---|
| Go API Gateway | Render Web Service | Auth, routing, audit, rate limiting |
| Python Intelligence | Render Web Service | Analytics, risk modelling |
| Frontend | Netlify CDN | Static web delivery |
| Secrets | Render env-var groups (generateValue) / Docker secrets in the air-gap stack |
JWT + engine secrets are generated, never committed |
# Backend — Render
# Push prexus-kernel to private GitHub repo → connect to Render
# Required environment variables on Render:
# ANTHROPIC_API_KEY → Your Claude API key
# OPENAI_API_KEY → Your OpenAI API key
# GEMINI_API_KEY → Your Gemini API key
# JWT_SECRET → Run: openssl rand -base64 32
# CORS_ALLOWED_ORIGINS → https://your-app.netlify.app
# Frontend — Netlify
# Drag and drop /frontend folder to Netlify
# Update API_BASE in meteorium.html to point to Render URL# Build
docker build -t prexus-kernel:latest .
# Run
docker run -p 8080:8080 \
-e JWT_SECRET=your-secret-here \
-e ANTHROPIC_API_KEY=your-key \
prexus-kernel:latest
# Health check
curl localhost:8080/health| Technology | Role | Version |
|---|---|---|
 |
API gateway, middleware, audit ledger | 1.22 |
 |
Analytics, risk models, IPCC integration | 3.11+ |
 |
Monte Carlo engine, VaR/CVaR numerics | 1.77+ |
 |
Structured API endpoints, persistence | 13+ |
 |
Predictive intelligence dashboard | 18 |
| CesiumJS | 3D globe, geospatial visualisation | 1.114 |
| Three.js | WebGL heatwave overlays | r128 |
| Cloudflare | Managed cloud hosting | — |
| Source | Domain | Cadence | Integration |
|---|---|---|---|
| Sentinel-1 SAR | Physical / Geospatial | 6-day revisit | ESA Open |
| ECMWF | Meteorological | 51-member ensemble | API adapter |
| IPCC-AR6 Database | Climate Scenarios | Built-in | Bundled pathways |
| Terminal Database | Financial Signals | 12 ms lag | API adapter |
✅ v1.4 Meteorium Engine — Physical risk scoring, Monte Carlo,
API Gateway · Meteorium UI with 3D globe
🧪 v1.5 Raksha Module — 360-degree clearance threat intelligence,
geopolitical risk modelling, sovereign operator dashboard
⬜ v1.6 Healtho Module — Population health risk engine,
bio-systemic shock propagation, epidemic modelling
⬜ v1.7 PostgreSQL Persistence — Full asset history, org workspaces,
audit-trail queryable database
⬜ v1.8 Real-time streaming — WebSocket push for live intelligence,
multi-asset portfolio event feeds
⬜ v2.0 Macro-Economic Module — Cross-domain risk correlation,
supply chain intelligence, geospatial signals
| Capability | Status | Module | Clearance |
|---|---|---|---|
| Health / Liveness Probe | ✅ Live | Core | Public |
| Organisation Registration | ✅ Live | Core | Public |
| JWT Authentication + RBAC | ✅ Live | Core | Public |
| Monte Carlo Simulation | ✅ Live | Meteorium | Level 2 |
| VaR 95% / CVaR 95% | ✅ Live | Meteorium | Level 2 |
| Tamper-Evident Audit | ✅ Live | Core | Level 2 |
| 3D Climate Globe | ✅ Live | Meteorium | Level 2 |
| Meto AI (Claude / GPT-4o / Gemini) | ✅ Live | Meteorium | Level 2 |
| Portfolio Aggregation | 🔄 Progress | Meteorium | Level 2 |
| PostgreSQL Persistence | ✅ Gateway (users, assets, applications) · ⬜ engine lake (SQLite) | Core | — |
| Raksha Threat Intelligence | 🧪 v0 scaffold | Raksha | ABAC in prexus_core |
| Healtho Risk Engine | 📝 Planned (no code) | Healtho | — |
| Macro-Economic Module | 🔨 Planned | Macro | Level 3 |
| Geospatial Signals | 🔨 Planned | Geo | Level 4 |
| Supply Chain Intelligence | 🔨 Planned | Supply | Level 3 |
| Real-time WebSocket Feed | 🔨 Planned | Core | Level 2 |
| Sector | Use Case | Key Modules |
|---|---|---|
| National Government | Climate resilience planning, infrastructure stress testing | Raksha, Geo |
| Central Banks | Systemic climate-financial risk, portfolio exposure | Meteorium, Core |
| Asset Managers | Portfolio-level climate VaR, regulatory disclosure (TCFD) | Meteorium |
| Insurance / Reinsurance | Physical risk underwriting, loss modelling | Meteorium |
| Infrastructure Planning | Asset optimisation, multi-scenario planning | Meteorium, Supply |
| Sovereign Wealth Funds | Long-horizon structural risk, geopolitical overlays | All modules |
Prexus Intelligence operates under a dual licensing model that cleanly separates open interface from proprietary intelligence.
The user interface, design system, and frontend components of Prexus are released under the Apache 2.0 License. This includes:
- All files under
/frontend/(HTML, CSS, JavaScript) - UI design tokens, component styles, layout system
- The Meteorium globe interface and dashboard shell
- Index, hub, landing, and demo pages
You may use, modify, and distribute these under standard Apache 2.0 terms.
The backend systems, intelligence pipelines, simulation engines, and data infrastructure are proprietary to Prexus Intelligence and are not licensed for external use, reproduction, or deployment without a signed agreement. This includes:
/backend/— Go API gateway, auth, audit ledger, risk proxy/data-engine/— Python intelligence layers (Layer 0–6), FastAPI endpoints/data-engine/rust/— Monte Carlo simulation engine, VaR/CVaR computation- All risk models, IPCC pathway integrations, scenario calibration logic
- The Prexus intelligence architecture, scoring algorithms, and data fusion methods
Commercial licensing, institutional pilots, and sovereign deployment agreements are available. Contact: dev@prexus.in
See LICENSE (Apache 2.0) and NOTICE (Proprietary terms) for full details.
P R E X U S I N T E L L I G E N C E
Sovereign Predictive Intelligence Infrastructure
For authorised institutional recipients only
© Prexus Intelligence. All rights reserved.