← RETURN TO DOSSIERFILE ELF.26 · CASE RECORD
EXHIBIT EDATA · PIPELINE — CASE STUDY

MARKET INTEL

An end-to-end Moroccan real-estate pipeline — scraping public listings, cleaning them, and detecting pricing trends — turning messy market noise into signals.

ORGANIZATION

INDEPENDENT BUILD

PERIOD

2024 — 2025

ROLE

END-TO-END: SCRAPING · CLEANING · ANALYSIS — SOLO

STACK

PYTHON · SCRAPY · PANDAS

01 — THE BRIEF

Moroccan real-estate data is public but useless in its raw form: scattered across listing sites, inconsistently formatted, full of duplicates and fantasy prices. Anyone trying to read the market is reading noise.

Market Intel is the full pipeline: collect the listings, clean them into a trustworthy dataset, and analyze that dataset for pricing and trend signals a decision could rest on.

02 — OPERATING CONSTRAINTS

  • Real-world scraping: inconsistent markup, duplicated listings, missing fields, dirty prices.
  • Cleaning is where the value lives — analysis over dirty data produces confident nonsense.
  • End-to-end ownership: the same engineer built collection, cleaning, and analysis.

03 — SYSTEM STRUCTURE

SCRAPY COLLECTORS

Python spiders harvesting public listings at scale.

CLEANING LAYER

Pandas normalization — deduplication, field repair, outlier pricing filtered before it can lie.

ANALYSIS & TREND DETECTION

Pricing and market signals extracted from the cleaned dataset.

04 — KEY DECISIONS

DECISION 01

Treat cleaning as the product, not a chore

The pipeline's credibility is decided before analysis starts — a trustworthy dataset is what separates signal from confident noise.

DECISION 02

Build the full pipeline, not a scraper

Collection alone produces files. Capture → clean → analyze produces answers — the same capture-to-decision pattern as the enterprise work in this file.

DECISION 03

Boring, proven tools

Scrapy and Pandas are unglamorous and battle-tested — the ambition went into the data quality, not the stack.

05 — OUTCOME

Messy public listings became pricing and market signals.

  • · The same domain's behavior data powers Exhibit F — the Reco Engine.

CROSS-EXAMINATION

The full walkthrough — code, schema, trade-offs — happens live. Fifteen minutes, no slides.

NEXT EXHIBIT

RECO ENGINE