01 The problem: a transaction with too many blind spots
Buying, selling, or investing in property in Kenya is a multi-party process that runs on fragmented information. A buyer has to discover a listing, judge whether it is fairly priced, confirm the seller actually owns it, arrange financing, and complete a transfer — often while working with agents, banks, and lawyers who each hold one slice of the picture. The questions that matter most are exactly the ones that are hardest to answer from inside any single slice: is this title genuine, is this price honest, and is this investment sound?
CHACADOM INVESTMENTS set out to build the platform that holds the whole picture: one ecosystem where every stage of the transaction lives, every stakeholder has a workspace, and the data — listings, documents, market signals — is structured so it can be reasoned about. My mandate was the reasoning layer: making that structured data usable as answers, scores, and verifications at every stage.
02 The constraints that shaped the build
- Trust is the product. An AI layer in a property platform cannot guess. Every answer the assistant gives and every score the platform displays has to be traceable to listing and market data — or users lose trust in everything else.
- Eight audiences, one brain. Buyers, sellers, landlords, tenants, investors, developers, banks, and institutions each need a different slice of the same truth. The AI layer had to be consistent across all eight workspaces, not a per-corner chatbot.
- Real-world connectivity. Users are on mobile networks, often on mid-range phones, sometimes on 2G. Responses had to be light and the platform usable on low bandwidth — the same discipline I apply on every Africa-facing build.
- Three languages, one voice. English, Kiswahili, and Français delivery — the assistant and the workspace copy have to work for users in all three.
- High-stakes documents. Titles and identity documents are where fraud attempts concentrate. Verification had to be designed as tooling for human decisions — structured checks and traceable evidence — not a black box.
03 The lifecycle: seven stages, one ecosystem
The platform organizes every property journey into a 7-stage transaction lifecycle. The AI layer exists to make each stage smarter:
04 What I built
"Ask Keja AI" assistant
A conversational layer grounded in the platform's listing and market data. Users ask in plain language — about a property, an area, or a process — and get answers tied to the actual records, not generic model text. It lives inside the workspaces, so it answers with the user's role in mind.
Per-listing investment scoring
Every listing carries an investment score computed from its own structured attributes and market context — so "is this worth my money?" stops being a guess and starts being a comparable, explainable number next to every property.
Title & document verification tooling
Structured checks that surface document signals for the platform's verification workflow, and feed the KEJA Trust Score. Designed for auditability: the outcome links back to evidence, so a verification is a claim you can inspect, not a score you have to take on faith.
Market data & trust infrastructure
Market context behind listings, the KEJA Trust Score visible per property — with 70 listings certified at Trust ≥ 80 as of October 2026 — and tokenized-asset support in the platform's invest stage, on top of the same verified data spine.
In high-stakes products, the AI layer is only as good as the data spine beneath it. Because scoring and verification draw from the same structured listing data the platform transacts on, an answer, a score, and a transaction record can't quietly disagree with each other. That coherence — not any single feature — is what makes the trust score mean something.
05 Results: what's live and measurable
| Dimension | Where it stands (Oct 2026) | Why it matters |
|---|---|---|
| Live platform | Public trial at keja.app — real listings, real users, real transactions pipeline | The system survived contact with production: it is not a demo. |
| Scale | 87+ listings across 12 Kenyan counties | The scoring and market-data layers hold up across markets, not one neighborhood. |
| Trust | 70 listings certified at Trust Score ≥ 80 | Verification tooling produces auditable trust signals — the platform's core currency. |
| Reach | English · Kiswahili · Français; mobile-first | The AI layer serves users in their language on the devices they own. |
| Recognition | Nominated for an Africa AI Future Award (real-estate AI category) — ceremony 27 Nov 2026, Nairobi | Independent recognition that the product is doing something worth noticing. |
Keja AI set out to make property transactions in Kenya transparent — and the AI layer Victor built is what makes that real in production. The Ask Keja assistant answers from our actual listing data, the investment scores give every property an explainable standard, and the KEJA Trust Score has become the number our users check first — 70 listings were already certified above 80 in the first months of public trial.
What this means for your business
- If you run a marketplace or vertical platform — logistics, agriculture, health, finance — the pattern transfers directly: structure the data spine first, then put answers, scores, and verification on top of it. Keja AI is that pattern applied to property.
- If you're sitting on untapped proprietary data — documents, listings, transactions, claims — that data is a product. Scoring and verification layers convert trust into something visible, and visible trust converts users.
- If you've been told AI can't be trusted in your industry — that's usually an architecture complaint, not a model complaint. Grounded answers, traceable evidence, and auditability are the difference between an AI feature and a system a regulator, bank, or buyer will accept.
Want this kind of AI layer in your platform?
Book a 30-minute AI systems assessment. Bring your workflow or your data spine — you'll leave with an honest read on what's buildable, what it takes, and the smallest useful first step.