Programme note / Product

Dubai Real Estate Valuation Toolkit

An instant Dubai property valuation where every figure names its source, and the range widens when the evidence is thin.

My role
Solo: the wizard, the five API routes, the pricing engine and its evidence-tier model, the hand-curated benchmark table with a source recorded per row, the saved-report surface, and the QA suites that drive the real browser rather than asserting on the markup.
Made at
BARNES Dubai
Built
2026
Status
In beta
Access
Public test build on dubai-real-estate-valuation-toolkit.vercel.app; nothing submitted reaches a broker
Stack
Next.js 16, TypeScript, Zod, Vercel, Turso (libSQL), Resend, Playwright
The Dubai Real Estate Valuation Toolkit landing page: a dark nav with the toolkit's name, the headline 'How much is your property worth?' over a monochrome aerial of Dubai's skyline and highway interchange, a line promising three questions and a figure with the evidence that sets its range, and a card headed Get your free valuation with an address field.

Chapter 01 of 04

The brief

An instant Dubai property valuation that shows its working.

A three-step wizard turns an address into a value range, and every figure on the result names its source and the month it was read. When the evidence is thin, the range visibly widens instead of hiding behind a number nobody measured.

The property portals already answer "what is my flat worth" in seconds, and they answer it as a black box: a figure, sometimes a range, and no way to see what produced it. For an owner that is a guess with a logo on it. For a valuer it is unusable, because the first question a professional asks of any number is where it came from.

The tool started as a page inside the BARNES Dubai innovation portal and was carved out into its own repo, then rebuilt from a plausible demo into something a valuer could read closely. The earlier version printed a confidence score and a range that contradicted each other, on a bar drawn at a fixed position. It looked like a measurement and was decoration.

[ Open the test build ]

Chapter 02 of 04

The problem

Give a Dubai owner an instant valuation where every figure on the screen says where it came from, and where thin evidence shows up as a wider range instead of a confident number.

Chapter 03 of 04

The build

Frontend Engineering
A three-step wizard (address, property details, contact) and a results screen in Next.js 16's App Router: a range bar drawn to scale from the band, an itemised price-adjustment breakdown, an evidence table, a trend-extrapolation panel, comparable sales, a twelve-month trend chart, and location signals from public open data (RTA metro stations, KHDA school inspections, OSM Overpass). The five result surfaces were extracted into shared components so the wizard and the saved report cannot drift apart.
Backend Engineering
Five API routes on Vercel. Estimate runs the pricing engine; lookup resolves a deed or Makani number; lead writes to Vercel Blob and notifies through Resend; email-report sends the result; report persists one to Turso and serves it back at its own URL. The report route recomputes server side and never accepts a client-supplied price, so nobody can publish a fabricated figure on a branded page. Every integration no-ops cleanly rather than erroring when its env var is unset, including the database: with no Turso URL the save button does not render at all, rather than rendering and failing.
Data Pipelines
A hand-curated benchmark table, 15 communities, where each row records the published index it was read from and the month. Automated ingest from the Dubai Land Department is closed off (its transaction search is captcha-gated per query and the bulk portal is unreachable), so curation with recorded provenance is the model rather than a stopgap. One row is flagged in place because its villa index does not state whether its per-sqft denominator is built-up area or plot, and the two differ enough to matter.
Product Strategy
The positioning is the valuation that shows its working. Bayut and Property Finder are black boxes; the differentiator is that every displayed number produces an evidence row inside the engine rather than in the page, which is what stops a figure reaching the screen without provenance. Where a number cannot be sourced the surface says so out loud rather than quietly deriving it from the figure it was meant to explain. Net yield is the sharpest case: it is the number the portals do not publish, and it needs a service charge, which is published in no feed at all. So the charges are curated by hand as a band per community, and the yield is published as a range, because the sources make a single figure indefensible: Downtown Dubai runs from AED 21 per square foot in one tower to 67.88 in another. Three of the fifteen communities have no cited source and therefore no net yield, which the evidence table states plainly.

Build notes. How it came together and the decisions on the record.

[ Read the full build notes ]

Chapter 04 of 04

What it did

A public test build where no figure reaches the screen without a source.

The four benchmark rows that once read 'unverified' now each cite a published index and month. Checked against the live production API in September 2026: a mid of AED 3,210,000 on a band of AED 2,889,000 to 3,531,000, exactly the curated-benchmark tier's plus or minus 10 percent.

The tool is deployed as a standalone prototype, the Dubai Real Estate Valuation Toolkit, with no BARNES branding and a banner saying plainly that nothing submitted reaches a broker. The live production API returns a band that matches its evidence tier exactly, and every figure on the result card carries a row naming its source and the period it covers.

15
Dubai communitieshand-curated, a source recorded per row
5
API routesestimate, lookup, lead, email-report, report
±6-15%
range widthset by the evidence tier, not by a score
  1. 01The product claim is that the valuation shows its working, so every figure names its source and the period it covers. The rows are built in the pricing engine, not the page, which is what stops a number reaching the screen without provenance. Where a figure cannot be sourced the row says so rather than going quiet.
  2. 02The range bar is drawn to scale inside a fixed frame, so a wider evidence tier visibly reads as a wider band. It used to be hardcoded at left 15 percent, width 70 percent: decoration that looked like a measurement.
  3. 03Every adjustment is itemised rather than folded into one unexplained number.
  4. 04Deliberately not drawn on the same chart as the twelve-month trend: both are heuristics, and stacking them into one picture would read as a measurement.
  5. 05The comparables and the trend chart say on their face that they are illustrative, because they are derived from the community benchmark rather than pulled from the transaction register.
  6. 06The methodology is a link away from the figure, not buried in a terms page.
  7. 07Three steps, with the tower name carried through to the estimate.
  8. 08The prototype banner is the first thing on the page, above the pitch.
Next on the programmeApplied AI / Feature / 2026
The BARNES Data Centre on a monitor at a working desk in daylight: the light interface with its rust-orange accent, a sidebar running from All Properties through to an admin group, a wide filter row across the top, and at the centre the bespoke radar empty state reading 'Search to load properties' above '836,270 properties on file'. A notebook, a pen and a handwritten viewing list sit on the desk beside it.

Data Centre. BARNES Dubai's internal data tooling: an entity-centric valuation platform and AVM, absorbed into a broker tool over ~840,000 ownership records.

[ Read the programme note ]

Contact

Pavlo Puzikov, a black and white studio portrait in a pinstripe suit.

Pavlo Puzikov

Marketing and innovation: immersive 3D, applied AI, generative content

Write to

pavlopuzikov@gmail.com

Open to
Freelance projects, remote. Full-time roles, onsite or hybrid in the EU.
Based in
Bulgaria
Most recent role
Marketing and Innovation Lead, BARNES Dubai, 2026. Product Innovation Researcher there from 2025.
Education
BCom, Marketing and Management, University of Auckland, 2020 to 2023
Languages
English, Russian