Determistic Modern Risk Platform. AI Orchestration.
Built on Google Cloud.
What is RaaS?
Consume governed risk states for both regulatory and economic risk analytics, with schemas, tolerances, and lineage baked in. Results stay explicitly separable across regulatory production and non-regulatory what-if analysis, while the deterministic kernel stays fixed and AI orchestrates mapping and explanation only.
Results contract
* Jurisdiction coverage is not uniform out of the box and is delivered via scoped module and overlay enablement.
Deterministic vs AI
Governed AI agents support intake guidance, mapping, readiness diagnostics, workflow support, and explanations. Pricing, Greeks, XVA, stress, SIMM/DIMM, SA-CCR/SA-CVA/Market Risk remain deterministic and versioned.
Join the roadmap poll to steer allowed vs. not allowed AI features.
Results Contract
Every result is versioned, tolerance-checked, and lineage-stamped. Confidence badges, Basel paragraph mapping, and deterministic kernels make regulators comfortable.
Structured JSON + PDF packs ready for API, BigQuery, dashboard, and regulator submissions.
Vannarho provides continuous, deterministic risk states with AI orchestration for explanations and mapping. Subscribe to results contracts with tolerances, lineage, and regulator-ready narratives.
Commercial
Pay for governed risk results with lineage and SLOs. Keep core access for exploration; add validation services when you need benchmarked proof.
Python wheels for teams exploring the deterministic kernel and APIs before subscribing to results.
Independent validation runs, benchmarking, and regulator-ready packs to prove the contract. We work alongside your teams to design future-state flows and set up the platform.
Subscribe to governed risk states (schemas, tolerances, lineage) with delivery via API, BigQuery, PDF, and console.
Results Catalog
Curves, pricing, dividends, XVA explain, sensitivities, stress, SIMM/DIMM, SA-CCR/SA-CVA/Market Risk power every risk state. Browse the governed scope and pull sample JSON/PDF.
Result Tiles
Each analytic becomes a result tile with schema, tolerances, lineage, and delivery options. Mix and match for trading desks, treasury, and regulatory teams.
*SIMM license not provided with Vannarho.
Capital & Regulator Contract
Counterparty Credit Risk (SA-CCR, IMM support), Standardised Credit Risk (SA-CR), IRB Credit Risk, Direct Credit Risk, Credit Risk Mitigation, CVA Risk (SA-CVA, BA-CVA, APRA CVA overlay), Market Risk (SA-MR, Simplified MR, IMA, APRA MR overlays), Output Floor/Aggregation Inputs (capital-floor component outputs).
IRRBB, CSRBB, Stress Testing Compute Support.
Quantitative Disclosure Feeds, Qualitative Disclosure Support, Disclosure Reconciliation Support.
Compliance as Code
Vannarho treats compliance as a build-time operating model, not a pack written after calculators and reports already exist. Regulatory meaning is formalized into stable objects, bound to code and outputs, and emitted back out as trace, audit packets, coverage, and AI-ready knowledge assets.
Normative truth
Stable normative objects remain the source of truth across refactors, schema evolution, overlay additions, and release-by-release audit comparisons.
Operational navigation
Reverse traceability becomes the regulator and audit walkthrough experience, but it is generated from bindings rather than hand-authored after the fact.
A canonical Basel core is formalized once, then regulator overlays tighten applicability, evidence, disclosures, labels, and effective dates without creating silent compute forks.
Rules bind to code hooks, config, tests, outputs, disclosures, scenarios, assumptions, and governance records so implementation, evidence, and releases stay connected.
Trace records, audit packets, coverage summaries, change-impact views, and disclosure walkthroughs are generated from the curated layer plus executed repo state.
AI consumes generated trace, evidence manifests, and regulatory knowledge objects. It does not invent compliance meaning that the deterministic system cannot prove.
Pillar 1
Deterministic rules, parameter tables, classification logic, and governed output bindings with release-gated evidence.
Pillar 2
First-class scenarios, assumptions, governance records, controls, and evidence-heavy audit packets.
Pillar 3
Disclosure bindings, reconciliation links, and explicit trace from published cells back to the governed upstream outputs.
Define products that inherit schemas, tolerances, explainability, and lineage. Keep AI out of pricing; keep custom logic inside a governed kernel with the same results contract guarantees.
Accelerate your model development and ensure accuracy with Vannarho as your powerful benchmarking tool. Validate proprietary internal models against our robust analytics engine.
Empower your Model Validation teams with an independent, transparent, and robust analytical framework.
Vannarho provides a sophisticated, primary pricing and risk model for institutions of all sizes.
Operating Model Outcomes
Ops reduction
35%
Less reconciliation and QA through deterministic outputs.
What-if speed
4x
Faster scenario turnaround with reusable risk states.
Team size
-25%
Clear governance boundary reduces support burden.
Cost Reduction
Access the full Vannarho user guide for the deterministic platform at the heart of the solution...for step-by-step instructions, reference details, and examples to get the most from the platform. This complements the examples and python wheels repository.
Download the full guide and explore the key sections that accelerate setup and ongoing use.
A fast refresher for quants and validators on the mathematical basis of the platform re market and credit risk, AI orchestration boundaries, and how we might evolve the kernel on Google Cloud. Not an academic paper—think roadmap and cross-links. We'd love your feedback!
What’s inside
Stochastic calculus, measure change/Girsanov, tensors/AAD on GPU, variance control, Sobol.
Basel III/IV mapping, governance, reg-to-code, IFRS 9 intersections.
MC/quasi-MC, copulas, PDE/FDM, adjoint maths, XVA sensitivities, DAG scheduling.
When ML wins, explainability, Bayesian/graph contagion, stress nets, auditability.
Streaming risk, state management, GPU/on-prem hybrids, observability, intraday capital/XVA.
Quantum gradients, neuromorphic/spiking nets, explainable-AI regulation, research roadmap.
Access our tools and repositories to seamlessly integrate Vannarho into your development workflows.
Get started with our official Python client library and explore practical code examples and tutorials.
Learn about Vannarho's architecture leveraging Google Cloud for scalable and secure risk management.
ExploreWhy Google Cloud?
With Google, Vannarho helps organizations build quickly, securely, and cost effectively with infrastructure optimized for AI, modern, enterprise, and distributed workloads.
This starts from their foundations: their terrestrial and subsea fiber has 10 times the reach of the next leading cloud provider; and their internal resource management and scheduling innovations, Borg and Colossus, enable us to run with nearly infinite scaling, performance, and low cost - benefits we ultimately pass on to our customers.
Google's AI-optimized supercomputing architecture, AI Hypercomputer, packages a decade of expertise to improve AI system performance and efficiency.
Their modern, container-based applications can scale higher than any of our competitors, and enterprise organizations can easily migrate critical workloads to the cloud and benefit from leading reliability & security.
Their distributed infrastructure is specifically designed to support the unique on premises requirements of sovereignty, scalability, and connectivity for organizations who have more stringent control requirements.
Customers choose Google because they're the only cloud with infrastructure designed from the ground up for the AI era.
Google Partner
Working with Vannarho as a Google Partner is not the same as going direct to the Google Cloud console. It adds escalation paths, optimization expertise, commercial flexibility, and outcome-specific delivery support.
1. Priority Support
We can escalate issues through Google partner support channels and reach Google engineers faster than a standard self-serve customer path. That means faster triage, better escalation discipline, and shorter turnaround on critical incidents.
2. Cost & Performance
We do not just provision infrastructure. We focus on FinOps, expensive repetitive operations, and architecture tuning so runtime performance and monthly spend improve together rather than trade off blindly.
3. One Bill
Where clients operate under Cloud Commit or minimum-spend arrangements, we can often structure delivery so software consumption and cloud consumption sit together commercially. That simplifies procurement and helps use committed spend more effectively.
4. Outcome-Based Competencies
The real value is repeated experience solving the same class of risk, control-plane, delivery, and governed AI problems. When needed, we can also reach into Google for deeper product or architecture support tied to that outcome.
Vannarho runs as a governed Google Cloud platform with a stable API boundary, a first-party workbench, dedicated tenant delivery surfaces, deterministic VRE execution, and curated serving and BI layers. Key Google Cloud components in the target operating model include:
Stable external API boundary for OAuth or JWT verification, quotas, request analytics, correlation IDs, and routing to canonical RaaS backends.
Delivers the first-party workbench and authenticates users before tenant-aware authorization and approval rules are applied inside the platform.
Hosts intake, translation, run control, worker glue, and projection services that turn submissions into governed control-plane objects.
Holds canonical registries, execution state, result facts, serving views, and BI marts that power APIs, dashboards, and client-facing analytical products.
Supports workspace context, collaborative drafts, notifications, and low-latency summary projections without becoming the source of deterministic truth.
Supports explanation, narrative generation, and guided operations within the deterministic boundary, never as the pricing or approval engine.
Provides infrastructure identity, secret custody, encryption control, and the security envelope that sits around every regulated surface.
Executes one resolved flow job at a time inside the deterministic compute plane and emits governed evidence and result artifacts.
Captures run diagnostics, lineage events, dispatch state, and support signals needed for evidence reconstruction and governed operations.
Consumes curated serving and BI layers for cross-role dashboards, regulatory views, and controlled downstream analytical consumption.
Forms the asynchronous control backbone for domain events, recurring schedules, retryable dispatch, and downstream projection refreshes.
Keeps shared and tenant surfaces connected through controlled ingress, tenant-specific networking, and private service reachability.
Provides immutable custody for raw submissions, bundles, evidence, result artifacts, and regulator-ready output packs.
Supports provider CI/CD for image builds, environment promotion, and controlled release automation around the platform.
Supports traceable runtime image publication for provider CI/CD and controlled deployment into the deterministic compute layer.
Additional implementation supports such as provider CI/CD, supporting delivery automation, and low-latency acceleration may still be used where justified, but the public product story centers on typed APIs, control-plane services, governed data custody, deterministic execution, and curated serving layers.
Workbench + Typed APIs
The workbench is the first-party client of the RaaS control plane, but it is not a bespoke backend. Client automation, operator tooling, and client AI agents all consume the same typed northbound APIs.
Current model
Governed Control Plane
RaaS turns client submissions into governed control objects with explicit lifecycle states. The platform owns intake, readiness, approval, scheduling, and delivery; deterministic VRE execution begins only after the run is resolvable and allowed to launch.
Portfolio, market, legal, and configuration inputs enter as governed submissions with validation issues, translation reports, and artifact custody.
Reporting scopes, config sets, rulebooks, and operational bindings stay explicit so regulatory context is queryable rather than hidden in runtime artifacts.
Approval policies, waivers, readiness evaluations, and schedule rules determine whether a candidate run can progress toward submission.
Only resolved and approved runs are dispatched. Evidence, result artifacts, load jobs, and serving readiness remain linked end to end.
That lifecycle matters because the product truth is not a single batch job. It is a set of typed objects, command-oriented transitions, evidence references, and serving projections that can be reconstructed for audit, client support, and regulator review.
Readiness & Config Doctor
RaaS adds a readiness layer above raw runtime syntax. Before launch, the platform checks whether the submission, bindings, approvals, and evidence state are structurally consistent with the requested run purpose and supported native seam.
Tenancy & Operating Model
The target production model is not a loose multi-tenant web app. It is a dedicated-project-per-tenant operating model with provider shared services, client-specific analytics and runtime surfaces, and optional advanced multi-geo patterns where data residency requires them.
Provider-managed control-plane, monitoring, billing support, and northbound ingress remain separate from client-owned delivery and runtime surfaces.
The default path gives each client dedicated `qa` and `prod` project boundaries for workbench, BigQuery, BI, retained data, and ephemeral runtime.
For stronger residency and cross-jurisdiction needs, geo-specific data projects can be introduced while only controlled outputs cross project boundaries.
Managed deployment assistance remains available where client operating models require it, but the reference architecture is a governed Google Cloud service with explicit provider-versus-client ownership boundaries and tenant-aware delivery surfaces.
Architecture
Clients, operators, and client AI enter through a stable API boundary and a first-party workbench. The platform resolves typed control objects, preserves evidence and artifacts, dispatches deterministic VRE jobs, and serves curated results through APIs, workbench views, BI, BigQuery, and regulator-ready packs.
Channels & API boundary
Client systems, operators, and client AI enter through Apigee plus the Firebase-hosted workbench with typed northbound APIs and tenant-aware sessions.
Canonical control plane
Intake, scope, config, approvals, readiness, schedules, and resolved runs stay explicit as typed control objects with lifecycle and audit references.
Evidence & custody
Immutable artifacts live in Cloud Storage while canonical facts, evidence registries, and output load state are registered into BigQuery-backed product truth.
Dispatch backbone
Pub/Sub, Scheduler, and Tasks coordinate recurring schedules, retryable dispatch, downstream refresh, and operational decoupling across the platform.
Deterministic VRE runtime
GKE hosts the regulated compute plane: curves, pricing, XVA explain, sensitivities, stress, SIMM/DIMM, SA-CCR/SA-CVA/Market Risk, and governed evidence generation.
Serving, BI, and delivery
Serving views, low-latency projections, Looker dashboards, APIs, workbench cards, BigQuery access, and regulator-ready PDFs all consume curated facts instead of raw files.
Delivery
Typed APIs over canonical facts, serving views, and evidence registries. Ideal for automation, embedding, and controlled downstream integration.
First-party workbench for lineage, movements, readiness, operational remediation, and guided explanations across the same typed APIs.
Query curated facts, serving views, and BI marts for regulatory, non-regulatory, and what-if analysis without coupling consumers to raw result files.
Regulator packs, audit narratives, and exec-ready summaries with mappings and tolerances.
We distilled the leading ideas from our AI research backlog. Cast your vote and tell us which innovation would elevate your risk practice the most.
Your feedback steers where we invest engineering time—from smarter explainability to faster onboarding and supervision tooling. We will share outcomes with the community.
Loading poll data...
See a demo with synthetic data. Then, when you are ready, get your results contract: risk states with schemas, tolerances, lineage, and regulator-ready narratives. We will run your data and share the outputs.
Schedule now