Gabriel Lima - Data Scientist & AI Researcher

About

A computer engineer passionate about statistics, artificial intelligence, big data, algorithms, cloud, and much more.

Federal Institute of Espírito Santo, Serra, ES

Pre-project: Towards a Transformer-based Architecture for Brazilian Portuguese Image Captioning
Research interests: deep learning, computer vision, natural language processing, multimodal machine learning, large language models, generative artificial intelligence.

Federal University of Espírito Santo, São Mateus, ES

University of São Paulo, São Paulo, SP

Mental Health Research:

Developed a script for patients segmentation with form-based data with clustering techniques, using Scikit-Learn and Python. Researched sentiment analysis through images, texts and audios. Researched diagnostic analysis based on socioeconomic data. Designed an architecture that makes use of machine learning and ontologies.

Federal University of Espírito Santo, São Mateus, ES

Smart Grids Research:

Developed a script for load detection in smart grids with Python. Applied dimensionality reduction in household appliance data with Keras, NumPy, Scikit-Learn, Pandas, Python, and Tensorflow. Collected voltage and (electric) current data from loads in smart grids simulated at the Laboratory of Renewable Energy I. Created a database with load data and its characteristics with MongoDB.
Designed a successful extraction of load characteristics with (electric) current data only. Simple estimator training with only the load characteristics with 98% accuracy (considering all classification metrics and the balance of the data).

Face Recognition Research:

Designed convolutional neural networks for face identification using with Databricks, Keras, OpenCV, Python, and Tensorflow. Developed scripts for extraction of facial features (embeddings) using convolutional autoencoders made with Keras. Designed a software for face recognition by the similarity of cosines between the faces feature vectors with NumPy and Scikit-Learn. Created a dataset using the loads data and its embeddings with NumPy files.
The final software results: 100% accuracy (also 100% precision and recall) achieved in specific cases and over 80% accuracy in another cases. Facial data extraction process takes less than a second, as well as storing and searching it.

Stefanini, Brasília, DF

As a consultant, I work closely with an international bank.

Tok&Stok, São Paulo, SP

Developed tables, views, partitions, and pipelines to feed the main tabular model from Tok&Stok, goals metrics and other destinations. Done with Analysis Services, AWS S3, Azure Automation, dbt, GitHub, Oracle, Pentaho, Power BI, SQL, and Snowflake.
Built market basket analysis to deliver insights about products organization and afinity, and continued sales forecasting project. Done with AWS S3, dbt, GitHub, MLflow, mlxtend, NumPy, Pandas, Python, SQL, Scikit-Learn, and Snowflake.
Built Slack bot features to generate alerts about Analysis Service database overloads, to publish Power BI reports by typing commands in Slack and to do chat actions. Done with APIs, AWS (S3, Kinesis, EKS, ECR, EC2, SQS, and Secrets Manager), Docker, GitHub, Kubernetes, Poetry, Power BI, and Python.
Helped maintain the data warehouse big data architecture and somes data pipelines. Done with AWS (S3, Glue, EC2), dbt, GitHub, Prefect, Python, Snowflake, and Terraform.

Tok&Stok, São Paulo, SP

Developed tables, views, and pipelines to feed some reports and an app to ranking sellers performance. Done with dbt, Oracle, Pentaho, and Snowflake.
Built reports and analyzed data to optimize decision making about customer budgets, sales performance, store performance, and product discounts. Done with SQL, Power BI, and Python.
Applied regression strategies to solve sales forecasting and to analyze goals metrics. Done with dbt, GitHub, NumPy, Pandas, Prophet, Python, Scikit-Learn, and Snowflake.