[ Back to Dubai Real Estate Valuation Toolkit ]

Build notes

Dubai Real Estate Valuation Toolkit: how it came together.

Develop · Build

How it came together.

  1. 01

    Provenance lives in the engine, not the page

    Every number the result shows produces an evidence row inside the pricing engine itself: the community benchmark, each percentage adjustment, the rent, the trend, the interest rate, each with its source and the month it was read. Because the row is built where the number is built, a figure cannot reach the screen without its provenance. Where a number cannot be sourced, the row says so instead of going quiet.

    Product Strategy
  2. 02

    Curation as the model, not a stopgap

    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 the benchmark table is curated by hand, each row naming the published index it came from and the month. One row stays flagged in place because its source does not say whether its per-square-foot figure is built-up area or plot.

    Data Pipelines
  3. 03

    Publish a range when the sources only support a range

    Net yield is the number the portals do not publish, and it needs a service charge, which appears in no feed. The charges are curated by hand as a band per community, so the yield is shown as a range. Communities with no cited source get no net yield at all, and the evidence table says so.

    Product Strategy
  4. 04

    A saved report cannot carry a made-up price

    A valuation can be saved and shared at its own URL. The report route recomputes the estimate on the server and never accepts a price from the client, so nobody can publish a fabricated figure on a branded page.

    Backend Engineering

Develop · Forks

Decisions on the record.

The few calls worth defending. Each one is a fork; the other branch would have been a different project.

  1. Decision · 01

    Every external integration fails open, not closed

    Pricing, deed lookup, the database, email and analytics are each gated behind an optional env var. Unset any of them and the tool doesn't error, it falls back: the curated benchmark table stands in for a live pricing service, demo data stands in for the deed lookup, and a lead is still saved to Vercel Blob with no email service configured at all. The wizard works end to end on a fresh clone with no keys. The rule extends to the UI, not just the server: the save-report card asks the server whether saving is on and renders nothing when it isn't, because a deployment with no database should show no button rather than a button that returns a 503.

  2. Decision · 02

    One tier-dependent number, not two

    The estimate used to carry two independent hardcoded ladders sitting next to each other: a confidence of 92/90/82/50 and a range of ±7/±9/±10/±15. Neither came from a measured error, and a reader given both would reasonably assume one was derived from the other. The fix was not to make the two agree but to delete one. A tier now carries exactly one number, the band half-width, which is a judgement ordered by evidence rather than a backtested error figure, and everything the reader sees about certainty is derived from it or is words rather than a number. A second tier-dependent constant appearing anywhere in the codebase means the bug is back.

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