Senior DBA at one of Spain’s leading online travel agencies. Database systems processing high volumes of bookings across a large accommodation catalog. Designed the search infrastructure behind the site’s autocomplete and suggestions – Redis Search over the MySQL catalog for full-text, geospatial, and fuzzy matching. Explored ML applications with the booking and review data on the side.

The Context #

In a travel booking platform, search speed and inventory accuracy are everything. Hundreds of thousands of accommodations, booking spikes during holidays, and databases that had to be fast and reliable under pressure.

The Work #

Performance architecture. Architected MySQL, MongoDB, and Redis clusters handling high booking volumes across the full accommodation catalog. Implemented automated failover for high availability during peak seasons.

Search and data integration. Built ETL pipelines integrating multiple booking providers. Implemented Redis Search over the MySQL catalog for full-text, geospatial queries, and autocomplete. Designed schemas supporting real-time inventory synchronization. Also experimented with Python (Keras/TensorFlow) for review sentiment analysis and booking pattern detection – early ML exposure, not production-grade.

Operations. Prometheus monitoring with proactive alerting. Query optimization across large datasets. Caching strategies to absorb traffic spikes without degradation.

MySQL (Master-Slave replication), MongoDB, Redis, Redis Search (RediSearch: full-text + GEO), Python, Prometheus/Grafana, Red Hat Linux.

Lessons Learned #

• Small teams that care outperform large teams that don’t. A handful of people willing to do whatever it takes – from infrastructure to search optimization – moved faster than any org chart would predict.
• Infrastructure can’t outrun the software. You can optimize databases all day, but if the application layer is inefficient, the bottleneck just moves. Performance is a full-stack problem.
• Curiosity pays off. The ML experiments weren’t planned – the data was there, and I wanted to see what was possible. Nothing production-critical came out of it, but it changed how I think about data.