SELECTED WORK

Projects built around real data problems.

A selection of work combining data mining, data quality, statistical modelling, machine learning and practical implementation.

Explore projects Discuss a project

Data Mining Data Quality Machine Learning Entity Resolution Statistical Analysis

THE WORK

From raw information to a result that can actually be used.

The projects below reflect the way I tend to approach data work: collect carefully, inspect quality, model only when useful, validate the output and keep the final result interpretable. They also show how different technical stages can be combined into a single end-to-end workflow.

FEATURED PROJECTS

WEB APPLICATION · 2026

SPECIALIST AREA

Web Entity Resolution Engine for SReality Listings

A browser-based implementation of my entity-resolution methodology, designed to make duplicate detection and graph reconciliation visible to a non-technical user.

Users can submit rental-listing URLs and activate a workflow that compares candidate records, estimates duplicate relationships and returns final entity groups after graph-based reconciliation.

Python R Random Forest Correlation Clustering Docker Firebase HTML / CSS / JS

Open live project Entity Resolution service

URL 01 SReality listing

URL 02 SReality listing

URL 03 SReality listing

↓

PAIRWISE CLASSIFICATION Random Forest

0.91 0.74 0.18

↓

GRAPH RECONCILIATION Consistent entity groups

INPUT Records

→

OUTPUT Entities

MASTER'S THESIS · 2024—2026

Improving the Quality of Information on Prices of Dwellings Advertised in Aggregator Platforms

An end-to-end study of duplicate rental advertisements as a data-quality problem in online housing-market analysis.

I mined and structured Prague rental listings from SReality, engineered pairwise similarity features, built and validated Random Forest classifiers, and transformed pairwise duplicate predictions into entity-level datasets through alternative graph-resolution regimes.

Python R Scrapy Random Forest Graph Analytics Pyomo GLPK

Discuss a project

BALANCED RESOLUTION

Duplicate resolution materially changed the observed stock.

RAW
LISTINGS 2,607

→

RESOLVED ENTITIES 2,377

STOCK CORRECTION −8.82%

The balanced correlation-clustering regime provided a middle ground between conservative and expansive graph-resolution assumptions.

INDEPENDENT PROJECT · 2023

Lisbon Residential Market Analysis

An automated data-mining, cleaning and statistical-analysis pipeline built from 7,005 residential sale listings covering 22 of Lisbon’s 24 districts.

I developed a sequence of custom programs to handle the workflow end to end: Mercury mined up to 20 attributes per property from iMovirtual; Themis repaired and restructured misplaced values; Veritas removed likely duplicate listings; and Gaea treated extreme observations before analysis.

The resulting final sample contained 4,056 residential units. A fifth program, Cadmus, automated descriptive statistics, assumption checks and inferential testing at both city and district level, executing 207 t-tests and 115 descriptive functions and generating the material for a 76-page analytical report.

At the Lisbon-wide level, the analysis found statistically significant positive associations between sale price per m² and amenities including air conditioning, elevators, parking, river views and town views. The cleaned sample had a median sale price of €500,000, with median net and gross areas of 86 m² and 95 m² respectively.

Python Web Scraping Data Cleaning Deduplication Outlier Treatment Inferential Statistics Automated Reporting

Data Collection service Data Analysis service

AUTOMATED MARKET-ANALYSIS PIPELINE Lisbon Residential Market

2023

MERCURY Mine the market 7,005 raw listings · up to 20 attributes each

↓

THEMIS Repair & structure 6,891 cleaned records

↓

VERITAS Remove likely duplicates 5,381 retained records

↓

GAEA Treat extreme observations 4,056-unit final analytical sample

↓

CADMUS Analyse & report City-wide + district-level statistical outputs

DISTRICTS 22

T-TESTS 207

REPORT 76 pp.

MEDIAN PROPERTY IN FINAL SAMPLE

€500k sale price 86 m² net 95 m² gross

Discuss a project

ACROSS THE PROJECTS

The common thread is not one tool. It is the way the problem is handled.

END-TO-END

Work across the whole analytical chain.

Collection, cleaning, feature construction, modelling, validation and final output are treated as connected stages.

DATA QUALITY FIRST

Do not assume the source data is already trustworthy.

Missing values, duplication, inconsistent structures and measurement choices are part of the analytical problem itself.

TRACEABLE OUTPUT

Keep technical decisions understandable.

Models and transformations should lead to a result that can be inspected, communicated and used beyond the code that produced it.

HAVE A PROJECT IN MIND?

Let’s turn the problem into something concrete.

Tell me what data you have, what is difficult about it and what you ultimately need to obtain. I can help shape the analytical workflow from there.

Discuss a project