How an AI tool is cutting land fraud in Kenya with smart verification

Land fraud in Kenya isn’t just a crime—it’s a betrayal of trust that shatters generations of progress. For many families, a plot of land in Nairobi’s outskirts isn’t just real estate; it’s a child’s future, a parent’s retirement, and a family’s only tangible legacy. Yet every year, thousands of Kenyans fall victim to a sophisticated form of theft called "air supply," where fraudsters sell plots that don’t legally exist, leaving buyers with nothing but forged documents and closed bank accounts.

The Lesedi Developers scandal exposed the scale of the problem when over Sh1 billion vanished into phantom estates across Juja, Ruiru, and Thika. Forensic experts later confirmed that many of these plots never existed on any official registry—yet the sales continued unchecked. The gap wasn’t technology. It was the inability to verify the chain of ownership, the physical location, and the legal validity of a plot in real time. That’s why one developer decided to build TitleTrust, an AI-powered verification system designed to expose land fraud before the money changes hands.

From personal loss to a tech solution

The creator of TitleTrust didn’t start out building fraud detection tools. Their journey began with a family tragedy: a relative lost their ancestral shamba after a forged title surfaced in court. The legal system offered no recourse. The deed was real-looking, the signatures appeared authentic, and the fraudster had already fled with the cash. That’s when the idea took root: what if the same technology powering chatbots could be trained to reason like a forensic auditor?

The turning point came in late 2025, when the team entered the Google Gemini 3 Hackathon. Gemini 3 Pro was the first AI model capable of deep chain-of-thought analysis at scale. Unlike traditional chatbots, it could trace the history of a land title, spot temporal inconsistencies, and flag documents that violated Kenya’s Land Act—all while explaining its reasoning step-by-step. The team realized they weren’t building a chatbot; they were building an autonomous investigation agent.

How TitleTrust uncovers land fraud in four steps

TitleTrust doesn’t just read documents—it interrogates them. The system processes every piece of evidence through a series of specialized agents, each designed to catch a different type of fraud. Here’s how it works in practice.

1. Upload the deal pack and start the investigation

A field investigator—often a community advocate or buyer’s agent—opens the TitleTrust mobile app and uploads the full set of land documents: the title deed, Green Card (registry history), Mutation Form (subdivision records), and Sale Agreement. Within seconds, the system generates a live timeline that tracks every action, finding, and verification step. This timeline becomes the audit trail that can be presented in court or shared with authorities.

The interface is designed for users without legal training. Instead of asking them to interpret conveyancing law, the app guides them through prompts like "Please take a photo of the beacon in the northeast corner" or "Upload a clearer scan of the survey map." The system handles the complex reasoning in the background.

2. The ForensicEngine detects legal inconsistencies

At the core of TitleTrust is the ForensicEngine (backend/forensic_engine.py), which uses Gemini 3 Pro’s reasoning capabilities to analyze land documents. It doesn’t just extract text—it cross-references every date, transaction, and party name against Kenya’s legal framework.

• Ownership timeline checks: Can a property be transferred before the seller legally owns it? The system flags any attempt to sell land that hasn’t been formally registered.
• Document chronology: A Discharge of Charge cannot predate the original Charge. Forgers often manipulate Green Cards to show this impossible sequence.
• Licensing verification: The Surveyor listed on a Mutation Form must be registered in the official gazette. TitleTrust checks this automatically.
• Area discrepancies: If the sum of subdivided plots exceeds the area of the original title, it’s a red flag for oversubscription fraud.

What makes this possible is Gemini 3’s "Thinking Mode," which forces the model to produce a visible reasoning trace. In legal disputes, that trace isn’t just helpful—it’s essential. An investigator, lawyer, or judge can review the model’s logic and verify why a document was flagged, not just that it was.

3. The GeospatialEngine confirms the land exists where it says

One of the most deceptive land scams involves the bait-and-switch: fraudsters lure buyers to a beautiful, accessible plot, then hand over a deed for a swampy, inaccessible parcel miles away. TitleTrust’s GeospatialEngine (backend/geospatial_engine.py) prevents this by validating the physical location of a plot against its legal description.

When an investigator visits the site, they capture GPS coordinates and high-resolution photos. The GeospatialEngine then:

• Compares the GPS trace against the parcel geometry from the title deed
• Runs plausibility checks, such as distance-to-boundary and location confidence scores
• Detects environmental clues, like riparian vegetation, that suggest the plot is in a protected river reserve
• Issues a clear verdict: either "Location verified" or "Location mismatch—please re-scan beacons"

If satellite imagery shows the plot is in a flood zone or a 30-meter riparian buffer renders 60% of the land unbuildable, the investigator knows to walk away before signing any documents.

4. The MarathonLoop keeps the investigation moving

Behind the scenes, the MarathonLoop (backend/agent/marathon_loop.py) acts as the system’s coordinator. It’s not just a task scheduler—it’s designed to mimic how a human investigator would approach a complex case.

• It advances the investigation through predefined stages without requiring user input
• It handles API failures with exponential backoff, retrying failed calls automatically
• It escalates only when human judgment is truly needed, such as when a photo is too blurry to read a beacon number
• It maintains session state even if the phone loses signal for minutes at a time in rural areas

The goal is to make the system feel like a knowledgeable assistant rather than a rigid tool. Users follow prompts; the system reasons. That shift in design turns a complex legal process into a routine field check.

The architecture behind a real-time fraud detector

TitleTrust’s engineering isn’t about flashy dashboards or buzzwords—it’s about solving a specific problem with the right tools at the right time. Every component, from the Flutter mobile client to the Python-based backend agents, was chosen for its ability to deliver real-time, defensible results.

Mobile client: Built with Flutter for cross-platform compatibility, the client maintains a deduplicated, sequence-aware timeline of all agent activity. This ensures that even if connectivity drops mid-investigation, the system can resume where it left off without losing critical data.

Backend agents: Written in Python, the agents (forensic_engine.py, geospatial_engine.py, marathon_loop.py) are modular and extensible. Each one focuses on a single responsibility, making the system easier to debug and scale.

AI integration: The system relies on Gemini 3 Pro’s native multimodality and 2-million-token context window to process full document sets, including images and PDFs, in a single session. The model’s ability to produce reasoning traces is what turns raw data into court-admissible evidence.

Data validation: All findings are cross-referenced against official registries, gazette records, and satellite imagery. The system doesn’t trust a single source—it validates every claim independently.

Looking ahead: a future where land fraud isn’t inevitable

Land fraud in Kenya thrives because the current system is reactive, not preventative. By the time a buyer discovers the forgery, the fraudster has already cashed out and vanished. TitleTrust flips that model by putting verification in the hands of buyers and investigators before the money changes hands.

The next phase involves integrating TitleTrust with Kenya’s national land registry APIs to enable real-time title checks. If successful, the system could become a mandatory step in any land transaction, reducing fraud at the source rather than chasing perpetrators after the fact. It’s not a silver bullet—but for thousands of Kenyans, it’s a lifeline back to the land they thought they’d lost forever.