Custom Precision Metric for CatBoost classifier with Optuna

Job Description: I am looking for someone with Catboost knowledge and who knows the ins and out of setting custom evaluation metrics. If you do not know CatBoost and Optuna please do not apply. What I am looking for is a custom metric for an unbalanced classifier with 3 classes: -1, 0, 1. The weight of the classes should be considered, however, the classes with valué are only the minority classes -1 and 1. The classes represent a gambling situation. TP for classes 1 gain $2.3 but TN lose $2.1,. For class -1, the gain is $2 but the loss is -$1.5. I am looking to maximise the total profit during optimization using Optuna. The objective function should maximize the gains of having choosing the Max amount of TP /(TP -+ FP) for classes -1 and 1 (there are no gains or losses for class 0). The objective should be multiclass, and the evaluation metric should be the custom metric in discussion. Task type is GPU. To summarize I would like to have custom loss metric and based on maximizing custom precision.. The model should be Catboost Classifier and Optuna hyperparameter optimizatuon. For someone who knows ML this project should only take a couple of hours. Deliverable is a jupyter notebook with made up data using sklearn.datasets.make_classification with n_samples=1000, n_features=100, n_informative=10, n_redundant=2, n_repeated=0, n_classes=3, n_clusters_per_class=2, weights=[0.07, 0.90, 0.03]. Please note that Im trying to eval the minority classes, in my example -1 and 1, but make_classification may return different labels. I want to establish a relationship with a good freelancer since I might have several small problems similar to this one.