Industrial Engineer, specialty in Robotics and Perception
After five years as an Automation and Test Automation Engineer, actively supporting multiple teams in CI/CD deployments, I am resuming my career from where I completed my master's degree, evolving and expanding my expertise in Machine Learning and contributing to the creation and training of neural network models based on transformers.
Rulebot 1st iteration demo
Rulebot started with the idea of creating a webapp helper to help people understanding specific rule sections from complex boardgames.
With Typescript used for the frontend and Haystack powering the backend, Rulebot effectively uses Gemma to generate a short explanations related to the rule snipet that the user question is related to.
During my Big Data Applications and Processes course, I developed an application to handle a portion of the large database from GDELT.
This applications allows you to work with large amounts of data without demanding lots of space thanks to parallel processing and continuous transformings, saving up to 70% of space when treating your data from GDELT !
Thanks to these efforts, me and my team we did a study about the media representation of the Palestine-Israel conflict through the GDELT database.
Check out the repository for this application here.
ASA - Lost object detector
My Automatic Surveillance Application was my Master's thesis project, which was formed by 3 different software applications : A Lost object detector powered by YOLO, a stampede dectector and a heatmap analysis tool.
Check out the repository for this application here.
Using the transformer library from Huggingface, we have fine-tuned a variation of the LLM Bert to predict what equivalent star ranting comes from customer reviews. Our fine-tuned model can predict with 71.1% of accuracy and up to Good/Great binary accuracy of 86.5 %, what punctuation from 1-5 values comes from customer reviews.
Among other Kaggle competitions project, the food classificator is a CNN that is based on a fine-tuned efficientnetv2-b0, achieves an average of 64% of Avg Accuracy its prediction, achieving a higher value than its counterpart based on Random Forest which achieved a 56.40% on Accuracy.
The projects you see here are only part of my recent work. In the coming months, I will be sharing more, including initiatives in data sourcing, refinement, and organization, along with new neural network models built on transformer architectures.