How DobValidator Uses AI to Verify Physical Assets On-Chain

Why Traditional Due Diligence Breaks at Scale

When we started Dobprotocol, we quickly realized that the hardest part of real-world asset tokenization isn't the smart contract — it's proving the asset exists, works, and generates revenue. Traditional due diligence relies on manual audits, PDFs, and trust in a single counterparty. That doesn't scale. Every new asset class requires a new auditor, a new process, and a new set of assumptions. Even worse, the results are opaque: you either trust the report or you don't. There's no middle ground for fractional investors who need continuous, verifiable proof.

We built DobValidator to solve this. It's an AI-powered agent pipeline that ingests raw data from physical assets — documents, photos, GPS coordinates, IoT sensors — and produces a verifiable trust score anchored on Stellar. The goal isn't to replace human judgment entirely, but to make the verification process repeatable, transparent, and cheap enough to apply to thousands of assets.

The AI Agent Pipeline

DobValidator runs a sequence of specialized AI agents, each responsible for a specific verification dimension. The pipeline is modular — we can add or remove agents depending on the asset type. For an EV charger, we run four agents; for a solar farm, we add an energy yield agent. Here's how they work.

1. Document Analysis Agent

Every asset comes with paperwork: invoices, warranty certificates, insurance policies, ownership deeds. The Document Analysis Agent uses a fine-tuned LLM (based on GPT-4 with custom embeddings) to extract key fields: asset serial number, owner name, date of purchase, warranty expiry, and more. It cross-references these against public registries where possible (e.g., DMV for vehicles, county records for real estate). The output is a structured document fingerprint — a hash of the extracted data — which is then compared to the operator's claimed data. Any discrepancy flags the asset for manual review.

We also check for forgery indicators: inconsistent fonts, metadata anomalies (e.g., PDF created after the claimed date), and missing signatures. The agent doesn't just read text; it understands document structure. For example, it can detect if a warranty certificate has been photoshopped by analyzing shadow consistency across the image.

2. Geolocation & Physical Existence Agent

An asset on paper is worthless if it doesn't exist in the real world. The Geolocation Agent requires the operator to submit a geo-tagged photo of the asset. Our agent runs a reverse image search to see if the photo appears elsewhere on the web (which would indicate stock imagery). It then extracts GPS coordinates from the photo's EXIF data and checks them against the operator's declared location. If the coordinates are within a reasonable radius (e.g., 50 meters for a charger), we consider the location verified.

But we don't stop there. The agent also analyzes the photo content: it uses a computer vision model to confirm the asset type (e.g., 'EV charger' vs 'parking meter'), checks for visible damage, and estimates the surrounding environment (indoor vs outdoor, urban vs rural). All this feeds into a location confidence score.

3. Revenue Verification Agent

This is the hardest part. How do you prove an asset earns money without access to bank accounts or payment processors? Our Revenue Verification Agent ingests transaction logs from the asset's payment system (e.g., Stripe, Square, or a custom API). It looks for patterns: number of transactions per day, average transaction value, and revenue consistency over time. We then compare this to the operator's projected revenue. If the actual revenue is significantly lower, the trust score drops.

For EV chargers, we also pull data from the charging station's OCPP (Open Charge Point Protocol) interface — kilowatt-hours delivered, charging session duration, and connector usage. This data is signed by the charger's hardware, making it tamper-resistant. The agent correlates this with payment data to ensure every kWh sold corresponds to a payment received.

4. Trust Score Aggregation

Each agent outputs a score between 0 and 1, representing its confidence that the asset meets the claimed criteria. The Trust Score Aggregator combines these using a weighted average. Weights are asset-class-specific: for a solar panel, revenue verification might be 40% of the final score, while for a data center, it's more about uptime and connectivity. The final trust score is a float between 0 and 1, anchored on Stellar via a hash of the raw data. This means anyone can verify the score by recomputing the hash from the raw data — no central authority needed.

On-Chain Anchoring via Stellar

Why Stellar instead of Ethereum or Base? Because we need low-cost, high-frequency anchoring. Each verification run produces a hash of the agent outputs (the raw scores, timestamps, and agent versions). We write this hash to Stellar's blockchain as a memo in a simple payment transaction. Cost: ~0.00001 XLM per write. This gives us an immutable timestamp and a public record that the verification occurred. Investors can query the hash and compare it to the published trust score on our dashboard. If they match, the verification is valid.

We also use Stellar's built-in KYC standard to link verifications to operators. An operator completes KYC once, and that identity is reused across all their assets. This reduces friction and ensures that a verified operator can't suddenly become anonymous.

Why This Matters for Tokenization

Without DobValidator, tokenizing a physical asset is a leap of faith. You rely on the operator's word and a static audit report that's outdated the moment it's signed. With our pipeline, investors get continuous, verifiable proof that the asset exists and earns revenue. Operators get a trust score they can improve over time — a kind of credit score for physical assets. And we get the ability to scale from 10 assets to 10,000 without hiring a thousand auditors.

The pipeline is already live for our first projects: EV chargers from eHive in Chile. We're onboarding solar farms and data centers next. Each new asset class requires tweaking the agents, but the core pipeline — document, location, revenue — remains the same.

If you're building in RWA tokenization, stop treating verification as a one-time checkbox. Make it continuous, automated, and on-chain. That's the only way to build trust at scale.