← MRM/Projects/Asset Governance Platform
Enterprise SaaS · Multi-tenant Live · public sandbox .NET 9 · React/TS · 12 languages

Asset Governance Platform

One multi-tenant platform to govern assets and regulated operations across many organizations — from a township to a hospital or a defense manufacturer.

Try it: agp.torsova.com — sign in with sandbox / sandbox. No signup, no email, no card.

Organizations that manage assets and regulated operations across many teams need one governed system, not a pile of spreadsheets or single-tenant silos. AGP is that platform: independent organizations register assets, run them through an enforced lifecycle, schedule inspections, route approvals, and control exactly who can do what — each tenant isolated, everything on per-organization dashboards with a complete, append-only audit trail. The same build serves a company at the Standard bar and a government site, hospital or defense manufacturer at an elevated one — the difference is configuration, not code — and it refuses to boot a production deployment that isn't configured safely.

401
anonymous GET /api/v1/assets — refused
200
GET /api/v1/sandbox/info — the control
PKCE S256
real Keycloak OIDC sign-in
0
consecutive hourly resets, no gaps at last check
0
application console errors on load
sandbox
username and password, both
agp.torsova.com — is it really running?
$ curl -si https://agp.torsova.com/api/v1/assets HTTP/2 401 anonymous read refused — this is not demo mode $ curl -si https://agp.torsova.com/api/v1/sandbox/info HTTP/2 200 the control: the request reaches the app, not the edge $ sign in sandbox / sandbox ✔ Keycloak OIDC at agp-id.torsova.com · authorization code + PKCE S256 ✔ tenant data wiped hourly · 28 consecutive resets, no gaps at last check ✔ 0 application console errors · 0 failed application requests on load

This is the live deployment, not a mock. Drive it yourself at agp.torsova.com with sandbox / sandbox — sample data only. Tenant data is wiped hourly, and an audit trail of what was done there is kept for seven days.

What is hard to copy
●ITAR enforcement that refuses, rather than warns — an ITAR-controlled contract cannot be awarded to a manufacturer with no ITAR registration on file, or a lapsed one. It is enforced in the award path itself, not surfaced as a dashboard warning. Above that sits an instance-level posture guard: if ITAR-controlled data exists anywhere outside the Government security profile, the platform refuses to start and names the offending tenants.
●A government oversight portal with three distinct roles — Contracting Officer, DCMA auditor and Inspector General each see a different, scoped view. The prime controls per-contract visibility with explicit toggles, and cost breakdown can never be exposed through them. Every government read is written to an append-only access log, so both sides know exactly when the government looked.
●Two-tier assurance, applied uniformly — every gate, certification and first-article record carries System Complete and Human Verified as separate states. The machine passing a check is explicitly not a person signing for it, and the distinction survives into the audit export.
●Material traceability from heat and lot to delivery — lot and heat numbers, country of origin and DFARS 252.225 status follow material through chain-of-custody to the delivered line item.
●AS9102 first-article inspection as a structured record with section-level sign-off — a rejection raises a corrective-action request automatically.
●Explainable risk scoring that is not machine learning, and says so — named constant weights, per-factor point contributions shown to the user, and an explicit no history ⇒ no score rule rather than a confident number invented from nothing.
●Regulatory impact analysis with no AI in the decision path — when a framework changes, the system tells you which manufacturers, contracts, certifications and material lots are affected, by deterministic rules, and it returns the rule basis it used. A compliance decision a buyer cannot audit is a liability, not a feature.
●Per-action permission re-derivation on the offline sync channel — field tablets queue work offline and sync with version-conflict detection; permissions are re-derived per action on arrival. This closed a real privilege-escalation path where a field role could otherwise drive a gate to Human Verified.
Asset lifecycle & registration
●Register physical or logical assets with RFID/barcode assignment, type classification, and placement in location trees and component hierarchies.
●Enforced five-state lifecycle — Registered → In-Service → Under-Inspection → Retired → Disposed. Invalid transitions are rejected at the domain layer, no matter who calls.
●Bulk operations, soft-delete & saved searches — multi-asset transition/delete with per-item results, a recycle bin with retention, and per-user named filters.
Inspection & verification
●Scheduled & recurring inspections with automatic task generation, pass/fail outcomes and an overdue-work dashboard.
●Evidence capture — photos and PDFs attached as validated files (magic-byte checked, size-capped, path-traversal stripped).
●Multi-channel notifications — in-app, email and SMS through swappable providers; a channel failure is logged and never breaks the originating request.
Approval workflows
●Configurable approval chains for asset transitions, budget changes and subcontract awards.
●Separation of duties, enforced in the domain — a requester cannot approve their own request, and a document's approver cannot be its submitter.
●Full decision history in the audit trail, with pending-approval dashboards per user.
Access control & multi-tenancy
●Deny-by-default RBAC — every use case is gated by an explicit permission across 33 named permissions; an unknown role grants nothing. An architecture test fails the build if a request handler is left unattributed.
●IdP-driven permissions — roles come from Keycloak / Entra ID; role→permission mappings are editable per organization without re-issuing tokens.
●Tenant isolation, hand-audited across 611 query sites — every entity carries an organization ID and every query filters by the caller's organization. Isolation is enforced by hand in each handler, so the September 2026 gauntlet enumerated all 611 DbSet query expressions and hand-read the 30 that lacked an OrganizationId predicate. A reserved platform tier holds permissions that no tenant role can ever be granted.
●Per-tenant dashboards & trends — day-by-day history, not just "now", with PDF/CSV export.
Regulated operations
●Earned value management to ANSI/EIA-748 — PV, EV, AC, SV, CV, CPI, SPI, EAC, VAC, overrun flagging and a program-level roll-up.
●Critical-path scheduling — a real CPM implementation with forward and backward passes, float and cycle detection, with baseline-versus-projected slippage and PDF export.
●Document revision control — bijective revision sequences (Rev A→B→…→AA), supersession records, controlled distribution, and separation of duties on release.
●Immutable contract communication log — a formal per-contract thread with acknowledgement, response-due tracking and escalation; entries cannot be edited after posting.
●Condition monitoring — sensor readings joined to the work and contracts attached to an asset, so "is anything late on my pump station?" is one question rather than two systems.
●Counterfeit-part controls — approved-manufacturer lists, quarantine and disposition, in the shape of AS5553 / AS6081. Reports are structured the way GIDEP expects them; AGP does not transmit to GIDEP, and there is no live feed.
●Regulatory alerts, honestly labelled — the impact analysis over an alert is deterministic and real. The alerts themselves are curated content, authored rather than scraped from a regulator.
Audit & governance
●Append-only audit log — every state change writes an entry (actor, action, entity, time, org). Full accountability for who changed what, when.
●Deployment security profiles — Standard / Government / Healthcare progressively raise the bar (mandatory KMS keys, longer audit retention, automatic logoff); a config doctor refuses to start Production if the chosen profile's requirements aren't met.
●HIPAA §164.312, clause by clause — a seven-row table keyed to the real subsections, all six technical standards covered, with the gaps that belong to the covered entity pushed there explicitly. For NIST 800-53 and DFARS / CMMC the claim is narrower on purpose: safeguards aligned to those baselines at family level, with the DFARS clause citations themselves real.
●A STRIDE threat model for the API — one summary table covering the API as a whole, plus a pentest self-assessment on file.
Beyond assets — operations & partners
●CRM, orders & finance — customers, an order lifecycle, PDF invoices, accounts-receivable aging, and dormant-account outreach. All money is exact decimal, never floating point.
●Subcontractor management — award and track subcontracts on an on-track / due-soon / overdue cockpit with transparent rules-based risk scoring.
●Scoped self-service portal — a token lets a customer, subcontractor or manufacturer see only their own records; cross-subject access returns 403.
●Machine-to-machine integration API — an external system of record upserts assets idempotently and receives HMAC-signed webhook events with durable retry, exponential backoff and a delivery log.
Global-ready & accessible
●Twelve interface languages — English and Arabic (right-to-left, mirrored via CSS logical properties), plus German, French, Norwegian, Swedish, Dutch, Spanish, Portuguese, Italian, Japanese and Korean. All twelve are at full 327-key parity, enforced by a test that fails the build if any locale drifts. That catalogue covers the asset, inspection, approval and CRM surface; the newer modules still hold about 482 hardcoded English strings, and no native-speaker review has happened yet.
●Accessibility work you can check — full keyboard navigation, a translated guided tour, per-route document titles, a colourblind-safe palette option and a dark theme whose contrast ratios are enforced by test. The honest limits are in the box below.
●Installable PWA — app shell cached; tenant data is never cached offline, because a governance system must always show live, authorized data.
●An idle session lock (NIST 800-171 3.1.10 / 3.1.11) that warns, then signs out.
Architecture
●Clean Architecture, dependencies inward — a framework-free domain layer holds the aggregates and invariants across 56 entity types; an architecture test fails the build if the dependency rule is broken.
●CQRS application layer — every operation is a command or query, wrapped by cross-cutting authorization, entitlement and validation that run before any business logic.
●Swappable infrastructure — the same code runs on an in-memory store for demo and PostgreSQL / SQL Server for production; encryption, audit, notifications, billing and storage are all ports with pluggable adapters.
●Hardened API + React/TS SPA — HTTPS/HSTS, per-tenant rate limiting, strict CSP, body-size caps, RFC 7807 errors; a React Query + OAuth2/PKCE front end across 30 feature modules and 35 pages.
Security & hardening
●Attacked in its own production instance. A scan of something that is not running tells you nothing, so the first job was making it run. Once the sandbox was live, two adversarial gauntlets on 14 September 2026 went at the real deployment at agp.torsova.com and produced 34 findings — 7 High, 14 Medium, 12 Low. Every finding was independently re-derived before it was accepted: two were refuted as measurement artifacts and several severities were corrected downward. All of the Highs are remediated, and the remediation is recorded.
●Real auth, really tested — a real-Keycloak OIDC integration test (Testcontainers) boots an actual Keycloak container, issues a signed JWT, and proves the API validates it, expands roles to permissions, and rejects tampered tokens. The same realm file ships for production.
●Encryption — TLS in transit; AES-256-GCM field encryption at rest with a fresh nonce per value. The key is injected by the deployment from KMS / Key Vault and is never committed; the config doctor refuses to boot production on a placeholder key.
●Tamper resistance — optimistic-concurrency tokens prevent silent overwrites; authenticated encryption makes altered fields detectable.
●Supply-chain discipline — warnings are build errors, NuGet audit runs at every build at the low threshold and fails it, and a generated SBOM ships with the release.
Proof
0
automated tests — 800 backend, 336 frontend. Two backend tests skip: one needs a live S3 bucket, one needs Docker. Both are environment-gated, neither is broken.
0 / 0
errors / warnings — warnings-as-errors solution-wide, and eslint --max-warnings 0 exits clean
0
lines of C# across 586 files
0
API endpoints across 46 controllers
0
entity types · 24 EF migrations · 33 named permissions
0
frontend feature modules across 35 pages
0
interface languages, all at full 327-key parity
0
query sites hand-audited for tenant isolation
0
findings from two gauntlets against the live site — every one re-derived, all Highs remediated
live
agp.torsova.com · sandbox / sandbox
ℹ
Readiness, not certification — stated plainly. The Standard / Government / Healthcare profiles ship technical safeguards aligned to baselines like NIST 800-53, HIPAA §164.312 and DFARS / CMMC. They make an audit straightforward — they are not a certification, and the platform cannot self-certify. Operating in a real government, hospital or defense environment still requires an independent outside audit or authorization (an ATO through an independent assessor, a HIPAA assessment, a C3PAO for CMMC) plus organizational work no software can do (SSP, POA&M, BAAs, DDTC registration, an AS9100 registrar, policies, training). What the platform guarantees mechanically: its config doctor refuses to boot an insecure production deployment, and it hard-enforces the gates it can. Built to be audited — not to fake it.
ℹ
Accessibility — the honest version. AGP publishes a VPAT 2.5 / Revised Section 508 Accessibility Conformance Report covering every WCAG 2.1 A and AA criterion, with the basis stated per row. It is a labelled self-assessment draft, not an independent audit. It records 11 "Partially Supports" rows, and it names its own largest gap first: generated PDFs are not tagged (no PDF/UA) — and those are exactly the artifacts sent to an auditor. Automated tooling catches only 30–50% of accessibility issues, and no assistive-technology testing has been done.
.NET 9 · C#Clean Architecture · CQRSEF Core · PostgreSQL / SQL ServerReact · TypeScriptReact Query · OAuth2 / PKCEi18n · 12 localesKeycloak / Entra ID (OIDC)AES-256-GCMTestcontainersNuGet audit · SBOM
🔒
The product itself is public — sign in at agp.torsova.com with sandbox / sandbox and drive it. The source, business logic and deployment configuration are private: this page is an in-depth capability showcase, not a distribution.

Running regulated ops
across many teams?

I build multi-tenant platforms that keep complex, regulated operations organized, access-controlled and provably auditable — and that are honest about where the software stops and an outside auditor begins. Open the sandbox first, then tell me what you need to govern.