Projects

Table of Contents

• Surgical Instruments Detection with Custom Trainning
• Patient’s Clinical Outcome Prediction
• Model aggregation to forecast reservoir production
• Car suspension design

Surgical Instruments Detection with Custom Trainning (2024 –)

• The objective was to automate inventory managment with computer vision.
• Reduced operator time by 99%. A task that previously took over 20 minutes is now completed in seconds using our AI app and phone camera.
• The model identified missing components. YoLo model was used with custom training.
• Some components (screws) were identified by their positions in the box, different clustering algorithms were used to split components in regions, rows and columns, so the exact missing component’s name was identified.
• The components and the containers were made of aluminium, which is reflective and yields low contrast. So I tested different light conditions, camera position and opencv morphological transformations. Some of the crazy tests and the solution’s evolution are displayed below.

(a) - Some strange results when I was playing with distortion correction, trying to fix the box's deformation, being sunken in the center, which made it harder to group screws in columns/rows.	(b) - Initial setup to control light conditions with LED, and detections with opencv.	(c) - Transformations in opencv to deal with the lack of contrast.
(d) - Great results using AI with a light panel below.	(e) - Achieved great results with AI in normal light conditions.	(f) - Detections with a segmentation model.

Patient’s Clinical Outcome Prediction (2023 –)

• Objective: Accelerate treatment for high-risk patients by leveraging real-time data.
• Developed, evaluated and monitored models for patient’s clinical outcome prediction. Deployed APIs for integration with the app.
• EHR (Electronic Health Record) data is continuously updated before the patient’s next scheduled consultation, potentially indicating risks in real-time. The model operates 24/7, delivering instant alerts to physicians and performing risk stratification, enabling faster interventions and improving patient outcomes.

Model aggregation to forecast medium-term reservoir production (2020 - 2021)

• Scientific initiation project, with a published paper in on of the biggest energy congress of the world, Rio Oil and Gas.
• I worked at Fortaleza’s research group, STORMS. We used realizations of a benchmark resevouir model “OLYMPUS” as models to forecast production.
• We associated weights to each model and tested different optimization alogirthms to minimize a developed loss function.
• Particle Swarm optimization and others Bio-Inspired optimization algorithms.
• This strategy was validated with an average prediction error below 5%;

Car suspension design - Apuama Racing - FSAE (2018 - 2019)

• Used mathematical models to create vehicle simulations (point-mass-model, biclycle-model).
• Used Tire Models to interpolate new data points from the tire’s data.
• Developed a Suspension distribution forces program in Matlab. We would use the maximum forces obtained in braking and cornering simulations to dimension and reduce mass of the suspension.
• Project to reduce steering torque for better handling of the vehicle.
• Method to measure the car’s center of gravity and validate design.