Skip to content
View datafromlopes's full-sized avatar
👨‍💻
👨‍💻

Highlights

  • Pro

Block or report datafromlopes

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
datafromlopes/README.md

Hi, I'm Diego Lopes 🤘🏻

Lead Data Engineer · Data & AI Architecture · MSc Researcher @ IME-USP · Text-to-SQL

I build large-scale data platforms and production AI systems, and I research natural language interfaces for databases.

At work I own the data and AI architecture of a global e-commerce (LATAM and Europe): event-driven ingestion on GCP, a medallion lakehouse on BigQuery, and a WhatsApp customer service agent built with LangGraph on Vertex AI, with RAG, MCP tool calling and evaluation pipelines. Before that I spent 7 years on distributed data systems, including a 12-node, 3-AZ Cassandra cluster handling 1.46B+ writes a day at 13µs P50 latency and 99.99% availability.

At USP my research sits between NLP and databases: how to fine-tune compact LLMs to translate Portuguese questions into executable geospatial SQL.

Website LinkedIn ORCID Hugging Face

Research

AtlasSQL-BR: A Brazilian Portuguese Geospatial Text-to-SQL Dataset with Spatial Hierarchies Diego O. Lopes, Kelly R. Braghetto. Proceedings of the 41st Brazilian Symposium on Databases (SBBD 2026), pp. 85-98. Paper · Code · Dataset

MSc in Computer Science, IME-USP (dissertation submitted, defense in October 2026), advised by Prof. Kelly R. Braghetto.

Interests

  • Natural language interfaces for databases and semantic parsing
  • LLM fine-tuning for specialized domains (PEFT: LoRA, IA³)
  • Execution-based evaluation of generated SQL
  • Portuguese and other low-resource settings in Text-to-SQL

Featured projects

Project What it is
atlas-sql-br Geospatial Text-to-SQL dataset in Portuguese, fine-tuning pipeline and execution-based evaluation framework (PostGIS, sqlglot, MLflow)
matrix_multiply_optimizer Matrix multiplication in C, from naive loop to cache-blocked and vectorized, profiled with gprof and perf. 99.3% less execution time than baseline

Stack

Languages Python · SQL · C Distributed data Apache Cassandra · Kafka · Apache Iceberg · Trino · Spark · Flink Cloud and platform GCP (BigQuery, Pub/Sub, Cloud Run, Vertex AI, Dataform) · AWS (S3, EC2, Glue) · Airflow · dbt AI engineering LangGraph · Vertex AI Agent Engine · RAG · MCP · PyTorch · Hugging Face Transformers · PEFT Databases PostgreSQL/PostGIS · Cassandra · MongoDB

Open to collaborating on

  • Text-to-SQL and natural language interfaces for relational and geospatial databases
  • LLM evaluation for specialized or low-resource domains
  • NLP datasets and benchmarks in Portuguese
  • Large-scale data platform architecture

Contact

📫 diego.lopes@datafromlopes.com · datafromlopes.com


⚡ Private pilot, aviation and aircraft systems enthusiast ✈️

Pinned Loading

  1. atlas-sql-br atlas-sql-br Public

    AtlasSQL-BR: A Brazilian Portuguese Geospatial Text-to-SQL Dataset with Spatial Hierarchies

    Python 1

  2. matrix_multiply_optimizer matrix_multiply_optimizer Public

    This repository provides optimized implementations of matrix multiplication algorithms in C. It explores techniques like blocking, vectorization, and loop reordering to maximize performance.

    C 1