Thomas Wuillemin

A REST-based data-server built to provide fast and secured access to raw customer data from the various business intelligence solutions sold by the company.

The company used an existing data server, but it could not match the requirements from the BI tools teams, and it was neither possible to make it evolve smoothly.

Besides leading the team that built the solution, I designed—from the drawing board—the VizApp data server to handle the following requirements:
- Modular to be adaptable to each customer with a microlith architecture
- Fast data delivery: As some queries in raw client data may take a few hours we added a caching layer—in fact, a tree of dependent caches— scheduled and security-aware
- Security aware, each customer had its own security rules, so a very configurable solution to accommodate almost any case. Also, we included a security-explanatory service to help teams perform the configuration, understanding the impact of their security rules.
- Modern technological stack: Java and Spring Boot.

GRANITE is a software dedicated to France's Special Operations Forces (BFST) that allowed them to gather, analyze, and communicate military intelligence on the field.

GRANITE was built as an integrated suite of tools on top of a SQL database. Although this was SQL-based, users' data were presented as objects, and the business data model included inheritance. The main tools of the application were the data visualization tools offering users multi-dimension analysis such as geographical (GIS) and temporal, causal (exploration of relations in the data's graph).

The primary hard point of the project was that the software was to be used (and was used!) in the field by small remote teams without IT support or even basic infrastructure.

The critical success factor was the deep involvement of end users in all project stages, from the first mockups up to QA. Some details may be too small to fit in a public contract but may strongly impact the usability of the final product and its acceptance.

I was the technical leader on the project, and I was also personally in charge of the interoperability tool.

My client was using Firebase (one of Google's NoSQL databases) to store all its business data. Although Firebase has some nice advanced features, like client-side caching, they were not used. Instead, Firebase was used to "mimic" a more classical relational database storage.

Used this way, Firebase has limitations such as no advanced query (hence no real BI), orphaned data, etc. Moreover, Firebase is invoiced by access (read/write of an object), and when the database stores millions of objects, prices tend to rise.

My work was:
- Make tools to analyze existing data models, in particular historical data.
- Detect inconsistencies and orphans and, in relation with the data-management team, either create an automatic fix, if possible or provide an easy way to fix them manually.
- From the analysis, create a standard SQL model that can host the data (the final model was around 156 tables).
- Make a migration tool.
- Update the back end.

An important part of my work was the coordination between the data management team, the business owner, and the UI team.

The migration was done in 2 hours and was a complete success matching all the objectives. In the long run, the client was very satisfied with the outcomes.