Portfolio: data, AI, machine learning and remote sensing projects

Esther

About

It's me again, Esther.

I work in operations, data and AI, and I am as passionate about technology as I am about literature. I design data first, AI and automation solutions that help managers, entrepreneurs and companies improve how they operate, including the least technical among them. Selected work below: dashboards, remote sensing and GIS, machine learning, AI and automation, with links, demos and case studies.

Operations, data and AI by day; novels, poetry and anime the rest of the time.

More about me →

Source code is not publicly displayed in this portfolio in order to protect confidentiality and intellectual property. Detailed technical documentation and selected code samples can be shared upon request during a professional discussion.

Machine learning · Remote sensing

Classi-fy: spotting oil palm from satellite imagery

The problem: knowing where oil palm plantations are is a precondition for tracking deforestation. In the DRC that information exists nowhere reliably, and the country is far too large to go and check everywhere.

The solution: a model that ranks plots by probability from free satellite imagery: it flags 90 plantations out of 100, and is right about 87 times out of 100 when it calls one. Evaluated on regions it had never seen, with maps of its hits and its misses.

Tools: Python, Google Earth Engine, scikit-learn, Sentinel-1 and Sentinel-2

Data quality · GIS

Field data quality and GIS workflow

The problem: a raw field collection file opens cleanly and says nothing about what is wrong with it. Duplicate identifiers, impossible dates, parcels claimed twice: all invisible until the map is already published.

The solution: a reproducible workflow that runs 22 explicit checks on every record, exports only what passes to the GIS, and sends the rest back with the reason attached. Verified by planting 37 errors on purpose: it found all 37.

Tools: Python, pandas, GeoPandas, Shapely, QGIS

Client delivery · AI

Security intelligence for teams operating in the DRC

The problem: a humanitarian security team has to decide, several times a week, whether staff can travel to an area. The answer is spread across field alerts, press coverage and years of reports.

The solution: a platform that gathers those sources into one picture, gives a 48 hour outlook for each zone with its confidence, answers questions from the organisation's own documents, and prepares five kinds of reports in French and English.

Tools: Web application, AI assistant, automated reporting

In production · Data and AI

Virunga National Park monitoring platform

The problem: activities, visits, media content, press coverage, land records and boundary surveys are recorded by different teams, in different places, and read by people who will never open a spreadsheet.

The solution: one platform that brings the seven together, updates itself every few minutes, watches 54 news sites, and writes briefs and reports in which the AI can improve the style but never the facts.

Tools: Web application, AI assistant, automated reporting

Product · Works offline

FieldReviewer, survey data checked before it goes into a report

The problem: nobody rereads hundreds of questionnaires one by one, yet one duplicate, one impossible date or one four minute interview is enough to put a wrong figure in a donor report.

The solution: a tool for people who are not data analysts: it runs twelve checks on any survey export, gives a verdict and a score out of 100, names the three things to fix first, and produces a nine sheet Excel workbook. No formula, no pivot table. It runs on your own computer, and your file never leaves it.

Tools: KoboToolbox, Google Forms, any Excel or CSV export

More projects are coming: I am rewriting them one by one, each with its case study.