Trusty RWA Assessment Engine

An AI-powered engine that turns fragmented real-estate evidence into an auditable readiness decision, risk register and remediation plan.

Comment

Inspiration

Real-world asset tokenisation often begins at the wrong layer.

Creating a digital token does not resolve unclear ownership, incomplete documentation, conflicting valuations, weak governance or missing investor controls. Before an asset can be tokenised responsibly, it must first become structurally ready.

As the founder of Trusty Digital, I have repeatedly seen that the main obstacle is not blockchain deployment. The real problem comes earlier: fragmented ownership evidence, incomplete documentation, unclear governance and inconsistent compliance preparation.

Asset owners and professional advisers still coordinate much of this work through email, spreadsheets, document folders and disconnected systems. Assessments are slow, difficult to audit and heavily dependent on manual interpretation.

I wanted to test whether our existing readiness methodology could be converted into a working AI-supported product rather than remaining a manual consulting process or internal framework.

We built Trusty RWA Assessment Engine, with Readiness Assessment as its first active module.

What it does

Trusty RWA Assessment Engine analyses a real-estate project evidence pack and determines whether the asset is sufficiently prepared to proceed towards compliant legal and digital structuring.

A developer, asset owner or analyst can upload documents in PDF, DOCX, TXT, Markdown, JSON or CSV format. The system then:

  • extracts and consolidates relevant project facts;
  • distinguishes confirmed evidence from unsupported statements;
  • identifies missing documents and unresolved information;
  • detects contradictions across the evidence pack;
  • evaluates legal, compliance, governance, operational and commercial risks;
  • calculates a structured readiness score;
  • produces a GO, CONDITIONAL GO or NO-GO operational decision;
  • generates a risk register and remediation plan;
  • assigns recommended actions, owners and target deadlines;
  • produces downloadable Markdown and printable HTML reports.

The assessment covers five control dimensions:

  1. Legal and ownership clarity: 25%
  2. Compliance and investor eligibility: 25%
  3. Data and documentation completeness: 20%
  4. Governance and operational readiness: 20%
  5. Commercial and lifecycle logic: 10%

The weighted readiness score is calculated as:

$$ R = 0.25L + 0.25C + 0.20D + 0.20G + 0.10M $$

A high numerical score alone is not sufficient. Critical issues, such as unverified ownership, unresolved legal authority or missing investor eligibility controls, can override the score and prevent the project from progressing.

How we built it

The application uses FastAPI as the backend and a purpose-built browser interface as the assessment workspace.

Uploaded documents are extracted and consolidated into a structured evidence pack. GPT-5.6 is used through the OpenAI Responses API to interpret information across multiple documents, identify inconsistencies and generate a typed assessment using Structured Outputs.

The output is validated through a Pydantic schema so that every assessment follows a consistent structure and can be rendered directly in the dashboard, risk register and final report.

We also built a deterministic control engine. This provides:

  • a stable no-key demonstration mode;
  • reproducible benchmark results;
  • a baseline for future model evaluation;
  • protection for non-negotiable blocking rules.

Codex was used throughout the build process for architecture review, implementation, browser testing, automated test development, failure-mode analysis, security verification, interface refinement and submission documentation.

The final public application was deployed as a FastAPI web service, with a public demonstration mode that does not expose an OpenAI API key.

Challenges we faced

The main challenge was not generating an impressive narrative. It was ensuring that the system remained disciplined.

In a regulated workflow, an AI system must not:

  • treat an unverified statement as confirmed evidence;
  • imply legal or regulatory approval;
  • conceal uncertainty;
  • bypass critical controls;
  • replace qualified professional judgement.

We addressed this through:

  • strict Structured Output schemas;
  • explicit separation of confirmed, missing and conflicting evidence;
  • critical-risk overrides;
  • mandatory lifecycle constraints;
  • deterministic benchmark controls;
  • professional-review boundaries;
  • action-oriented outputs instead of generic AI commentary.

We also encountered practical engineering challenges. Browser uploads can submit an empty multipart file object even when no document is selected. This initially blocked the intended no-key demo workflow and had to be identified, corrected and covered by automated tests.

The first live GPT-5.6 request was also blocked by API quota. After restoring the API balance, we repeated the verification successfully without exposing the key or raw API response.

Another challenge was making the result useful to different participants. An asset owner needs to understand what is blocking progress. A lawyer needs to see missing legal evidence. A compliance specialist needs eligibility and jurisdictional risks. A project manager needs owners, actions and deadlines.

The final workflow brings these perspectives into one operational assessment.

What we learned

The most important lesson was that AI creates greater value when it converts expert methodology into a repeatable operating process.

The model should not replace professional judgement. It should make evidence, uncertainty, risk and next actions more structured and transparent.

We also learned that deterministic rules and model reasoning work best together:

  • deterministic rules protect non-negotiable controls;
  • GPT-5.6 handles interpretation, cross-document analysis and context;
  • typed outputs make the result operational rather than conversational;
  • automated tests preserve the integrity of the assessment workflow.

The deterministic benchmark returned 50/100 and NO-GO. The live GPT-5.6 assessment returned 25/100 and NO-GO, applying a stricter interpretation of the available evidence. The different scores were acceptable because both systems identified that the project should not progress without remediation.

Accomplishments

During Build Week, we created:

  • a working document-to-decision application;
  • an end-to-end readiness assessment workflow;
  • structured cross-document analysis with GPT-5.6;
  • contradiction and missing-evidence detection;
  • critical-risk decision overrides;
  • a structured risk register;
  • owner-based remediation actions;
  • downloadable and printable assessment reports;
  • a public no-key demonstration mode;
  • a live GPT-5.6 Structured Output workflow;
  • automated regression testing for the core decision logic;
  • a public GitHub repository;
  • a publicly deployed web application.

The final automated test suite contains nine passing tests covering:

  • deterministic scoring;
  • blocking ownership risks;
  • browser-style empty uploads;
  • API health;
  • report generation;
  • professional disclaimers;
  • invalid mode handling;
  • missing API-key behaviour.

The fictional demonstration project correctly identifies:

  • unverified ownership evidence;
  • missing SPV documentation;
  • absent independent valuation evidence;
  • unresolved encumbrance information;
  • governance gaps;
  • conflicting asset values.

What is next

Readiness Assessment is the first active module of a broader Trusty RWA Assessment Engine.

The next development stages include:

  • evidence-level citations for every finding;
  • human approval workflows for legal and compliance specialists;
  • CRM and secure data-room integrations;
  • project-specific readiness benchmarks;
  • Maturity Assessment for projects already entering implementation;
  • continuous monitoring after project approval;
  • generation of a verified digital Asset Passport;
  • expansion from European real estate into additional real-world asset classes.

The longer-term objective is to create an infrastructure layer covering:

  1. Readiness Assessment
  2. Maturity Assessment
  3. Continuous Monitoring
  4. Verified Asset Passport

Trusty RWA Assessment Engine is not designed to replace lawyers, compliance professionals, valuation specialists or investment decision-makers.

It is designed to give them a consistent operating environment in which evidence, uncertainty, risk and responsibility are transparent from the beginning.

Built With

Share this project:

Updates