
Case study 01
May 2026
Fraud Risk Intelligence
An end-to-end fraud analytics platform built on the IEEE-CIS dataset, combining BigQuery, dbt, LightGBM, FastAPI, and a live executive dashboard.
Verified project library
Selected projects lead the default view. Supporting industrial and experimental work stays available as a compact archive without competing with the current data-focused positioning.
Filter the review path
Role fit
Status
Evidence
Domain
Filter changes are preserved in the page URL.
Last verified 8 October 2026

Case study 01
May 2026
An end-to-end fraud analytics platform built on the IEEE-CIS dataset, combining BigQuery, dbt, LightGBM, FastAPI, and a live executive dashboard.
Case study 02
September 2026
A production-shaped forecasting system over 804.5M rows of market microstructure, built on a 16 GB laptop, where the main result is an investigation into why the leaderboard disagreed with the hold-out.
Case study 03
September - October 2026
A value-at-risk and expected-shortfall application built on a fork of Goldman Sachs' open-source gs-quant library: it measures the risk of a series or a portfolio and then tests whether the VaR model deserves to be trusted, with every forecast judged on the return that followed it.
Published evidence
VaR backtested with Kupiec, Christoffersen and Basel traffic light

Case study 04
June–August 2026
A predictive-maintenance analytics case study on 10,000 AI4I records, combining EDA, engineered sensor signals, evaluated failure-risk scoring, DuckDB SQL, and a ranked maintenance queue — plus a second, genuine RUL-regression case study on NASA C-MAPSS turbofan data.

Case study 05
June 2026
A steel-defect classification case study using 1,800 NEU-CLS images, handcrafted visual features, model benchmarking, error analysis, and an entropy-ranked review queue, extended with a measured embedding-RAG assistant and a live 3D digital twin of the inspection line.
Case study 06
September 2026
A retinopathy grading study on 3,662 fundus photographs that tests every claim it makes: whether the model reads the retina or the camera, whether it holds on two other populations, and - when it failed on one - whether the cause was the model or its labels.
Case study 07
September - October 2026
Two live Kaggle competitions worked end to end: Enveda CASMI 2026, a featured competition that ranks candidate molecular structures for MS/MS spectra, and Playground S6E10, a tabular ROC-AUC competition on airline satisfaction. In both, decisions were made on cross-validation and the public leaderboard was treated as a noisy sample.
Case study 08
February 2026; revised October 2026
An explainable anomaly dashboard for industrial (OT) telemetry: a statistical detector decides, an LLM only explains the verdict, and a labelled rule-based fallback takes over when the LLM is unavailable. The detection method is evaluated offline on the public BATADAL attack benchmark, with a modest result reported as such.
Published evidence
BATADAL: 43 signals, 14 attacks, one false-alarm budget for all detectors
Case study 09
June 2026
A football analytics platform built on Transfermarkt data with BigQuery, dbt, Power BI, and scikit-learn market-value modeling, developed in a fork of an earlier collaborative project.
Case study 10
April 2026
A tested analytics-engineering and BI portfolio spanning dbt/BigQuery marts, DuckDB CI, Power BI, DAX, Olist analysis, cohort/RFM SQL, and statistical testing.
Freelance solutions · delivery example
A small, fixed-scope Python and SQL-friendly workflow for BI and software agencies that need dependable data preparation and reporting support. This is a public, synthetic demonstration, not a client engagement.
Excel/CSV Reporting Automation
Clean and merge inconsistent files, quarantine invalid records, calculate agreed KPIs and hand off reusable Python code with validation tests. A live HTML dashboard shows the sample output.
Verified synthetic sample
200
Input rows
192
Valid rows
8
Flagged rows
9
Automated tests
Source row reconciliation passes. CI runs on Python 3.11 and 3.12.
01
Import
Two synthetic sources: sales CSV and customer Excel workbook.
02
Validate
Identify invalid values, conflicting orders and unmatched customers.
03
Reconcile
Check that accepted and excluded records account for every source row.
04
Deliver
Export clean data, quality logs, KPI tables and a standalone report.
The hosted page is a static report; it does not accept client file uploads. Scope, fees, confidentiality and delivery dates are agreed after reviewing anonymized sample data.
Discuss a scoped pilotCapability map
Projects where raw business data is transformed into reporting-ready tables, dashboards, and reviewable decision surfaces.
Projects where the modeling layer is framed around prioritization, risk segmentation, or operational review rather than generic benchmark claims.
Applications where a language model does the work: embedding retrieval measured against a keyword baseline, answer faithfulness scored by NLI against a no-retrieval control, multimodal document parsing, and AI-assisted triage.
Projects where model evaluation, class-level error analysis, and targeted human review matter as much as the visual signal itself.
Archive work that demonstrates industrial workflow design, multimodal interface thinking, and operator-centered product packaging.
Archive work that keeps tracking, storefront instrumentation, and GA4/GTM context connected to the broader analytics portfolio.