profile image

Eduardo Lima Barros

Artificial Intelligence engineer | LLM | Fine-tuning | Machine Learning | Python Back-End

Contact Me

About Me

I began my IT journey at 13, working on computer repairs and software setups with my brother. By 14, I was studying Computer Science at the Federal Institute, where I delved into algorithms and programming. After 5 years, I pursued a law degree to grasp business intricacies, later working as a consultant to help companies navigate risks. My consultancy experience, combined with launching my own business, equipped me to understand tech company challenges. Post-graduation, I worked in São Paulo, analyzing data systems and providing strategic insights on big data and governance.

That's why moving on to develop ML and AI models was just a natural step in my career. Experience with systems, management, programming, IT, data and governance has given me the perfect tools to contribute to the field. In this way, I can use this knowledge to help companies implement machine learning models to make better decisions and solve business problems with AI.

Which is what I've been doing for the last few years. Training models whose data source is the customer bases I map, integrate, clean and pre-process the data and results of the bases to include in the training and generate valuable information for any manager.

Latest Projects

EasyFeeling: Sentiment Analysis Made Easy
New

EasyFeeling: Sentiment Analysis Made Easy

I fine-tuned Llama-2 model is capable of rapidly and accurately detecting positive or negative sentiment in new text data. By employing techniques like QLoRA and quantization, we've optimized training efficiency without compromising performance. Using a dataset accessible in the Stanford University, this model demonstrates adaptability to diverse sentiment analysis applications. I fine-tuned Llama-2-7b-chat-hf using QLoRA, Adam, Peft, bitsandbytes for quantization, and Hugging Face, achieving training completion in approximately 40 minutes on the free Google Colab platform.

Back my project
YOLO-CV-Cattle-Detection

Real-Time System Detection

Real-Time Detection System using AI and computer vision - Implemented with YOLO for object detection and SORT for tracking, ensuring robust and accurate video processing. Enhancing with a video frame manipulation module using Kalman Filter and DSA for efficient animal tracking.

Find out more

Luigi.AI

Luigi.AI - The perfect AI assistant for your own data

Luigi.AI — Innovative app that combines advanced techniques like RAG and technologies like Langchain, OpenAI API, and Streamlit to provide intelligent insights from your documents. Simply upload files (PDF, docx, txt), ask questions, and receive accurate answers.

Find out more

Sentiment Analisys

Text classification with Social Networks

Classification and prediction of discussions on Reddit, using web scraping and API and supervised learning, applying algorithms such as KNN, Random Forest and Logistic Regression from sklearn, TF-IDF-Vectorizer and TruncatedSVD for Natural Language Processing, and sampling the results with Matplotlib and Seaborn.

Find out more

project name

Mario Bross

Using Python and reinforcement learning to make the AI learn to play Mario and learn from each mistake. Using OpenAi GYM, Pytorch, NesPy and the PPO model.

Find out more

Work Experience

Software Engineer - LGPD2U/FortSecure (2023 - present)

Medium-sized company specializing in Information Security and Privacy, with more than 600 active clients in 10 countries. Promoted in less than 6 months to mid-level.

    Data and AI:

  • Integration of structured and unstructured client databases via data streaming: Use of Docker, containers and Kubernetes to manage infrastructure.Development of Webhooks and RESTful APIs for interaction with other Python and JS systems. Creation of customized datasets for each client.

    • Model training:

  • Use of the vendor's pre-trained model. Unsupervised training to identify patterns in client data. Adjusting the model with specific data from the client's databases, using supervised training with tagging and classification techniques, if necessary.

    • Generating insights:

  • Developing customized dashboards to visualize results. Analysis of results to identify opportunities for improvement.

    • Software development:

  • Developing a solution with JS and Sales Force and Workflows to identify agents responsible for storing customer data on personal devices, which resulted in a reduction in complaints and an increase in customer compliance.
  • Developing a tool to automate the deletion of customer data, alleviating the manual workload of the marketing team.

Systems specialist - Peck Advogados (2021 - 2022)

Large law firm in São Paulo, specializing in law and technology, working for major national and international clients.

  • Analyzing systems, applications, and APIs to determine and identifying security and compliance flaws in application codes, for clients such as Nike, Bimbo and SBT TV.

IT manager - C&T (2015 - 2016)

Medium size accountant firm

  • Managing IT infrastructure and network servers — Linux and Windows.
  • Monitoring threats and applying security measures.

Other Projects

Diabetes Prediction using Sage Maker Friendly

Using the model XGBoost to predict the probability of a pacient develop diabetes. Deploying on AWS Sage Maker and generenating end point for using in others aplications

Find out More

Neural Network From Scratch Friendly

Neural Network from scratch - A Deep Learning introduction

Find out More

GPT from Scratch Friendly

Gentle introduction to Generative Pre-trained Transformer

Find out More

More on GitHub