Ramses Benzari | Data & Business Analyst

Case study · Automated pipeline

The US data job market, tracked every day.

I built this live dashboard to monitor the US data job market, support my own job search, and potentially help others who are building a career in data. The pipeline collects new data job postings every day, uses AI to extract and enrich 16 fields per posting, from tools, skills and industry to salary and work setting, and publishes the results to a live eight-page Power BI dashboard.

PythonBigQuery SQLGemini APIPower BIGitHub Actions

Live dashboard ↗View code ↗

Data Scientist pays far more than any analyst title

Median advertised salary by role, for the ten most-posted roles. Analyst titles sit within a narrow $88K–$102K band.

Data Scientist is also the most-posted role in the dataset.

Only 15% of roles are fully remote

Share of postings by work setting.

15%fully remote

On-site 13.4KHybrid 5.1KRemote 3.3K

SQL and Excel remain the core requirements

Postings that mention each tool.

SQL and Excel each appear in about one posting in three.

The most jobs are not where the highest pay is

Top 10 cities by number of postings, coloured by median salary. New York and Washington DC lead on volume; Sunnyvale, Mountain View and San Jose lead on pay.

$95K–$115K$115K–$140K

Map of the top 10 US cities by number of analyst job postings. New York has the largest circle, followed by Washington DC, which is in the higher salary tier.

Automated daily pipeline

Runs every day on GitHub Actions with no manual steps. New postings go through the same cleaning rules automatically.

1 · Collect

JSearch API

13 analyst searches, six result pages deep, every morning.

2 · Extract & enrich

Gemini API

Turns each free-text posting into 16 structured fields: tools, hard and soft skills, industry, education, salary, work setting, benefits and more.

3 · Clean

BigQuery

Two SQL views remove duplicates, standardise locations and assign role categories.

4 · Report

Power BI

Eight-page dashboard, refreshed daily and published online.

Running daily since 26 May 2026

Every cleaning decision was measured, not assumed

Each posting has two possible sources for the same fields: the job API and the AI reading the description. I checked a random sample by hand to see which source filled each field correctly, then built the pipeline to take every field from the source that does it best.

100% vs 53%

Share of postings where work setting and education were filled: 100% from the API, only 53% from the AI, because many descriptions never state them. So these two fields come from the API, and the AI handles skills, tools and industry, which the API does not provide.

1,467 → 28

Free-text industries grouped into 28 categories, covering 98.8% of postings.

203

True duplicates removed by using the posting URL as the key, with no valid rows lost.

2 silent breaks

Made the pipeline resistant to API changes

The pipeline broke twice because of changes on the provider's side, with no error raised. First, the API retired its endpoint and runs reported success for eight days while collecting nothing. Then I found its job IDs changed on every call, so one posting had been stored eleven times. I made the posting URL the primary key and added a daily record count, so a provider update can no longer break the data silently.

Known limit: the same job posted on two different platforms has two URLs, so it is still counted twice. That cannot be controlled from my side.

6 sources → 1

Verified the data against an outside benchmark

The first version used six free job APIs and collected about 100,000 postings within days, more than national labor statistics suggest exist. The AI assistant helping me reported no issue. I inspected the raw table, found heavy duplication, and replaced the six sources with one reliable paid API. Data quality improved immediately.

Case study · Investment analysis

Where does an Airbnb actually pay off?

If you bought a 1–2 bedroom apartment to rent on Airbnb, which city would give the best return? I cleaned Inside Airbnb listings for 123 cities, matched them with property prices and utility costs from Numbeo, converted 35 currencies with exchange-rate data, estimated occupancy, and ranked 80 cities by annual return after running costs.

PythonDuckDB SQLpandasInside AirbnbNumbeo

Read the full analysis ↗

All 80 cities, by return on a city-centre purchase

Every top-20 city is in North America or Western Europe. That is partly a real result and partly the data: coverage is thin in Africa and Asia, and the Middle East is not covered.

World map of 80 cities coloured by return, from red (loss) to green (about 12%). The green cities are mostly in the US, Canada and the UK; most European and Asian cities are red or orange.

Top 10 cities by annual return

Where you buy inside a city can matter as much as which city. In Columbus, the return goes from 9.7% in the centre to 19.8% outside it.

Buy in the city centreBuy outside the centre

Cheap property only pays if people book it

Each dot is a city. Below 100 booked nights a year, no city returns more than 3.5%, however cheap its property.

7%+ return3.5–7%0–3.5%Loses money

0.81

Occupancy predicts return best. Its correlation with return is 0.81 on a scale from 0 to 1, against 0.36 for nightly price.

3 cities

lose money wherever you buy in them: Oslo, Hong Kong and Munich. Eight lose money on a city-centre purchase.

60 m²

Every city was costed on the same apartment size, since no source covers 80 cities. I re-ran it at 85 m²: cities moved 1.2 places on average, and the top 20 by one place at most.

The top 20: 17 in North America, 3 in Europe

12 US cities, 5 Canadian, plus Edinburgh, Barcelona and Valencia. Chicago leads on a city-centre purchase at 11.5%. Outside the centre, New Orleans leads at 22.5%.

Map of the 20 highest-return cities: 17 across the US and Canada, with Chicago the darkest green, plus Edinburgh, Barcelona and Valencia in Europe.

Why the size assumption favours North America

Every city was costed as a 60 m² apartment. Real apartments in the US and Canada are larger (about 68–102 m² in the US), so their purchase cost is understated and their return overstated. Dense cities like Paris or Hong Kong, with smaller apartments, get the opposite effect.

Limited coverage

The data only covers cities that Inside Airbnb's volunteers publish. Africa has two cities, and South Asia and the Middle East have none. This is not a ranking of every city in the world.

30nights a year

Validation

The occupancy model knew nothing about local laws. For Amsterdam, it landed exactly on the legal limit.

Occupancy was estimated from review counts alone. Amsterdam came out at 30 booked nights a year, which is the city's legal cap on holiday rentals. That is strong evidence the method measures something real.

Nightly prices were also checked against commercial data. Edinburgh: $332 here, against $311 (AirROI) and $283 (AirDNA).

About

Business sense, analyst's rigor.

I managed three concurrent construction projects for government clients, worth $2M combined. I led requirements gathering with architects, contractors and client stakeholders, built the cost estimates and budgets in Excel, and reported spending variances to management. I also launched the company website and its marketing, growing inbound enquiries from zero to about ten a month.

Before that, I spent nearly six years as a freelance architectural visualizer, working remotely with more than 50 clients in architecture, real estate and interiors, with a 60% repeat-client rate.

Both roles taught me what I now apply to data: understand the business question first, communicate clearly with every stakeholder, and make sure the numbers hold up.

Project Manager2023 – 2026Global Years · Government construction projects, Morocco
Architectural Visualizer2018 – 2023Freelance · 50+ clients, fully remote
Bachelor of Business Studies2022United International Business Schools

Areas of expertise

Data AnalysisBusiness AnalysisExploratory Data AnalysisData CleaningData ValidationData ModelingETL & Data PipelinesBusiness IntelligenceDashboard DevelopmentData VisualizationKPI ReportingStatistical AnalysisWorkflow AutomationAPI IntegrationRequirements GatheringBudgeting & Variance AnalysisStakeholder CommunicationProject Management

Certifications

Google

Data Analytics

Google

Advanced Data Analytics

Google

Business Intelligence

Microsoft

Power BI Data Analyst

Toolkit

Analysis

PythonpandasNumPystatsmodelsscikit-learnJupyter

SQL

BigQueryDuckDBCTEs

Spreadsheets

Microsoft ExcelPivot tablesXLOOKUP

Visualization

Power BIDAXPower Querymatplotlibseaborn

Data pipelines

ETLGoogle CloudGitHub ActionsGit & GitHubREST APIsGemini API

Languages

EnglishFrenchArabicSpanish (basic)

Contact

Let's find out what your data is really saying.

Open to Data Analyst, BI Analyst and Business Analyst roles, remote or in the US.

Connect on LinkedIn ↗ GitHub ↗