Ivan Ramon Jardon Rodriguez

OBJECTIVE
Developed a predictive model to forecast the remaining cargo capacity, known as underload, by predicting the following variables:

• Passenger forecast
• Baggage forecast
• Fuel forecast

APPROACH
The task required predicting continuous values, making it a regression problem.

MODEL
XGBoost is a powerful machine learning algorithm known for its effectiveness in regression tasks. It uses boosting techniques to create a robust model by combining multiple weak decision trees, enhancing prediction accuracy.

PROCESS
After training and fine-tuning individual models, a cascading ensemble was implemented. This approach combined the predictions from each model, resulting in a more reliable and accurate final forecast than a single model could provide.

BENEFITS
• Improved planning and reduced delays caused by limited cargo capacity, ensuring more efficient resource utilization.
• Maximized cargo capacity offer, unlocking a potential $2.5 million monthly profit.

OBJECTIVE
Developed and deployed a machine learning model in Google BigQuery to categorize customer satisfaction survey comments, enhancing feedback analysis and customer experience strategies.

APPROACH
Leveraged Gemini Flash for text classification, identifying key themes and detractor reasons affecting net promoter score (NPS).

PROCESS
• Collected and preprocessed survey responses.
• Built a text classification model using Gemini Flash.
• Deployed the model directly in BigQuery for scalable and efficient processing.
• Created a weekly tracking system to monitor top detractor reasons.
• Quantified NPS point loss per detractor category to prioritize improvements.

IMPACT AND BENEFITS
• Enabled analysis of customer feedback at scale.
• Provided automated tracking of detractor reasons and emerging trends.
• Allowed proactive mitigation of key issues, leading to potential NPS score improvements.

OBJECTIVE
Developed and deployed a phonetic similarity detection model to identify and mitigate potential confusion between flight numbers in air traffic management, enhancing operational safety.

APPROACH
Implemented phonetic similarity analysis using the Levenshtein distance and Python libraries like FuzzyWuzzy and Metaphone to quantify and rank flight number similarities.

PROCESS
• Collected and preprocessed flight schedules, focusing on departures and arrivals within one hour.
• Developed a Python-based model to calculate phonetic similarity scores between flight numbers.
• Automated detection of high-risk flight number pairs to prevent miscommunication.
• Designed a recommendation system for airlines to validate proposed flight numbers before assignment.

IMPACT AND BENEFITS
Enabled a scalable and automated analysis for ongoing monitoring and recommendation of flight numbers, reducing the risk of operational miscommunication by detecting similar-sounding flight numbers.