2026-04-13T06:10:28Zhttps://recercat.cat/oai/requestoai:recercat.cat:2117/4450422025-10-31T01:33:22Zcom_2072_1033col_2072_452951
RECERCAT
author
Masip Plasencia, Eduard
2025-10-31T01:33:22Z
2025-10-31T01:33:22Z
2025-10-23
http://hdl.handle.net/2117/445042
This thesis addresses the architectural limitations of Tauniqo, a growing SaaS company, whose purely transactional (OLTP) infrastructure impeded critical data analysis. This created severe business blind spots, including reactive customer management, uncontrolled AI operational costs, and an engineering team bottlenecked by manual data requests. The system's normalized schema was unsuited for analytics, leading to query times exceeding 250 seconds and risking the stability of the user-facing application. To solve this, a dimensional Data Warehouse (DW) was designed and implemented, chosen for its ability to provide complete performance isolation and empower non-technical users with self-service analytics. This V1.0 platform was built pragmatically using PostgreSQL, Python/Pandas, and Metabase to accelerate time-to-value. The project delivered three key outcomes: Enabled Data-Driven Customer Success: An automated dashboard reduced query times for key KPIs to under 3 seconds, shifting the CS team from a reactive to a proactive, data-informed approach. Established an AI FinOps Framework: The platform provided first-ever visibility into 78% of the company's AI costs, enabling data-driven optimization of LLM usage. Created a Strategic AI Dataset: An analysis-ready "ground truth" dataset was engineered, providing a foundational asset for systematic LLMOps experimentation to improve AI evaluation accuracy and cost-effectiveness. The core conclusion is that the primary performance bottleneck was the logical data model, not the underlying hardware. By successfully delivering immediate business value and fostering a data-informed culture, the project provides a validated foundation and a clear business case for a V2.0 roadmap to evolve towards a more scalable, enterprise-grade architecture.
Open Access
Àrees temàtiques de la UPC::Informàtica::Sistemes d'informació::Bases de dades
Multidimensional databases
Business intelligence
Data warehouse
Business intelligence
ETL
OLTP
OLAP
Dimensional modeling
Data pipeline
SaaS
Performance optimization
Self-service analytics
FinOps
AI
LLM
Cost observability
Data governance
Customer success
Scalability
Magatzem de dades
Intel·ligència de negoci
Analítica d'autoservei
IA
Governança de dades
Èxit del client
Escalabilitat
Escalabilitat
Escalabilitat
Escalabilitat
Escalabilitat
Bases de dades multidimensionals
Intel·ligència de negoci
Data-driven SaaS: architecting a dimensional data warehouse for BI and MLOps at Tauniqo
Master thesis