The initial project (v1/v2) was developed by SQLI, a French IT services company, with a custom PHP framework and Smarty templates. The legacy code contained 398 SVN revisions covering 2010-2013.

When I joined SQLI, 5 full-time PHP developers were dedicated to the project. Over time, I took on a growing share of the technical scope, eventually handling the entire perimeter on my own - a role that naturally evolved with my expertise and the autonomy that came with the PME environment. After covering the full perimeter solo for a while, I asked for reinforcement and obtained the hiring of a second developer.

From 2014-2016, our new Medialeads team undertook a complete rewrite of the platform, progressively migrating from the custom framework to Symfony 2/3 while maintaining the public site on a lighter custom MVC framework.

The platform was multilingual (7 languages), multi-country, with high product data volumes, and the need to maintain compatibility with existing supplier data feeds - all hosted on fully self-managed OVH dedicated servers: we handled absolutely everything ourselves, from system configuration to deployments.

• Massive data management & SQL optimization: I sharpened my mastery of SQL normal forms (1NF, 2NF, 3NF, BCNF), every join type (INNER/LEFT/RIGHT/CROSS/SELF) and advanced indexing (B-tree, composite, covering indexes), with constant `EXPLAIN ANALYZE` plan analysis, relying on MySQL master-slave replication and Memcache caching. I kept nightly batches of several hours running to recompute stock levels, prices and promotions across the whole catalog - at that scale a missing index could turn a 50 ms query into a 30-second timeout, so every millisecond saved per query compounded over millions of executions.
• Search engines & indexing: I went hands-on across four generations of search engines - (1) MySQL LIKE / FULLTEXT (MyISAM/InnoDB), (2) PostgreSQL full-text search with `tsvector`/`tsquery`, GIN indexes and `to_tsvector`/`ts_rank` functions, (3) Apache Solr, where I indexed hundreds of thousands of products with advanced faceting (categories, colors, prices, suppliers), spell-checking and custom relevance scoring, (4) Elasticsearch with inverted indexes, multilingual analyzers (stemming, tokenization, normalization), TF-IDF scoring, aggregations and facets. This journey let me grasp in depth the differences between a relational engine and a dedicated search engine.
• Complex algorithms & business rules: I designed and implemented numerous complex algorithms and intricate business rules - for example computing every possible combination of price, product variants and availability (Cartesian product / combinatorial explosion generating thousands of combinations per product across sizes, colors, quantities, marking types, marking zones and finishes). Dynamic pricing rule modeling and business decision engines are now part of my toolbox.
• I consolidated my mastery of the Symfony 2/3 ecosystem at production scale (Propel, Doctrine, bundles, services, security, WSSE).
• On legacy code migration, I built strong skills by driving the transition from the SVN monolith to Git-based microservices without downtime.
• On the business side, I gained deep domain knowledge of the B2B promotional products market, data feed automation and multi-country e-commerce.
• Finally, this project changed my way of working: I learned to coordinate inside a small team where each developer owns significant portions of the codebase, and to carry perimeters end-to-end.

With today's standards and hindsight, here are the choices I would now make differently - each decision made sense within the constraints of its era:

• Adopting a single framework from the start rather than maintaining a custom PHP framework alongside Symfony, though historical context justified coexistence during migration.
• Setting up CI/CD from the GitHub migration (January 2016) - Jenkins, Travis CI or GitLab CI were available even if GitHub Actions did not exist yet.
• Using Docker for dev and deploy environments, though Proxmox was the mature virtualization standard for PHP PME at the time.
• Centralizing data access via a single API rather than direct MySQL from each service - a pattern that only became widespread after 2017.