About
Me

Design, deployment, and scaling of a proprietary hybrid recommendation system, integrating automated data pipelines and specialized Deep Learning models for cultural analysis.

• Recommendation Infrastructure: Architected a scalable 'in-house' recommendation solution using FastAPI for the service API and Qdrant as a vector database for high-performance similarity search.

• Orchestration and MLOps: Implementation of robust data pipelines on Airflow for real-time ingestion and processing. Integration of MLflow for experiment tracking, model versioning, and algorithm lifecycle management in production.

• NLP & Sentiment Analysis: Fine-tuning language models (LLMs) for specific lexical recognition in cultural reviews. Developed a sentiment classifier optimized for the cinematic field, allowing for accurate global polarity extraction from user reviews.

• Personalization Algorithms: Development of hybrid recommendation engines combining User-Based and Content-Based (embeddings) approaches to offer ultra-personalized suggestions based on member history and cultural affinities.

• Infrastructure Optimization: Tailor-made technical architecture designed for SensCritique's needs, ensuring low response latency and smooth integration of new content and user feedback.

Development of tailor-made AI solutions and creation of innovative web applications for various clients and sectors.

Development of AI solutions and data pipelines for SEO predictive analysis, aimed at automating performance analysis and optimizing client strategies.

• ML Pipeline Development: Created an automated ML pipeline with Vertex AI for SEO performance predictive analysis. Designed and deployed algorithms (Random Forest, XGBoost) to predict organic traffic evolution, reducing client analysis time from 1 day to 30 minutes.

• ETL & NLP System: Designed an intelligent ETL system with NLP classification for automatic search intent detection on heterogeneous SEO data. Used K-Means semantic clustering to automate query segmentation, removing 95% of manual errors.

• Monitoring Models in Production: Implementation of drift monitoring via RMSE, MAE, R², and MAPE metrics. Integrated an Isolation Forest model for anomaly detection, increasing model reliability in real-world environments.

• Optimization and Automation: Developed an automation script for GCP project creation and data integration, reducing client onboarding time from 1 day to 30 minutes.

Development of web and mobile applications for the creation of innovative products in the accounting sector.