Edwin Isac

Achieved an mAP score of 0.89 by training brand data using TensorFlow 2.0 object detection API for a Brand & Celebrity Detection project, optimizing the model to enhance detection accuracy and reduce false positives. Improved a Reverse Image Search algorithm's accuracy rate to 92% by reducing images to a feature matrix using SVD and developing similarity measures. Managed and optimized Grafana, Kafka, and Zookeeper clusters for a program collecting data, enhancing system stability and reducing downtime. Created efficient and scalable web crawlers for data collection, achieving a 99% data extraction rate. Enhanced speech transcription model performance by training a HuggingFace transformer model and implementing noise reduction techniques. Deployed a program for extracting text from ad images with an accuracy rate of 98%, suggesting improvements for the OCR engine.

Developed an LSTM model to predict a person's pose using accelerometer and gyroscope data from smart devices for a Heterogeneity Activity Recognition project. Executed several successful Time Series Analysis projects including predicting patient admissions with 92% accuracy, predicting microbusiness density with 88% accuracy, and analyzing air quality patterns using the Temporal Fusion Transformer. Developed a tailored ad display system using predicted gender and age information, leading to significant increases in customer engagement. Developed scalable and reliable APIs for multiple models using Flask.

Blended the statistical knowledge with advanced computing techniques, learning to leverage ML algorithms and data visualisation tools to drive insights from complex data sets

Gained foundational knowledge in statistical theories and methodologies.