Diana Minine Gudinho

About

ML/NLP engineer focused on large-scale data linking for e-commerce-style catalogs and customer records, combining transformer embeddings (BERT/Sentence-BERT), NER, and FAISS-based vector search. Has delivered measurable lifts (e.g., +30% matching accuracy, Precision@10 62%→84%) and built production-grade, scalable pipelines in Airflow/PySpark with strong data quality and schema-drift handling.

Key Strengths

• Built NLP pipeline to unify multi-vendor product catalogs using BERT/DistilBERT + NER + FAISS
• Improved product matching accuracy by ~30%
• Designed scalable entity resolution with hybrid blocking + BERT similarity; scaled to tens of millions of records
• Improved entity match accuracy by ~25% vs prior system
• Improved semantic search relevance via Sentence-BERT fine-tuning; Precision@10 from 62% to 84% and ~35% relevance lift
• Production-grade data workflow engineering (Airflow, PySpark, Docker, CI/CD, monitoring, data quality checks)
• Handled vendor schema drift with automated schema validation and dynamic mapping layer
• Built and deployed a RAG-based compliance/legal document review system
• Designed for high-precision retrieval to reduce compliance risk
• Optimized retrieval and model performance for large-scale data (latency/compute constraints)
• Hands-on Airflow orchestration for ETL + ML pipelines at scale (5M+ daily records)
• End-to-end pipeline design from S3/Glue/PySpark to Redshift with DAG-based scheduling
• Structured agent/workflow testing approach (unit + integration) with measurable metrics
• Production monitoring with real-time dashboards and stakeholder feedback loops
• Effective collaboration with non-technical stakeholders (marketing/compliance) translating requirements into usable outputs

Experience

Education

Skills

Languages