OUR PROCESS

Training Data Collection

Data preparation is a crucial step in manifesting an efficient machine learning model.

A robust data collection process is established to ensure quality and accuracy of the training data. For example, we would specify photographing conditions for how training images need to be taken to ensure consistency with real life settings in object detection.

Data Labelling

We analyze the preliminary data collected for a better understanding of the context. Splitting it into 3 subsets for developing a legitimate model: training set, validation set and testing set.

We then work with the client to label the training set with custom defined objects, specifying the definitions for each type of object to achieve accuracy.

AI Modelling

Multiple models will be trained to determine the best specification on the grounding algorithm, data augmentation, feature selection and hyperparameters. Each and every model is evaluated with objective and precise metrics appropriate to the model.

Generating the Trained Model

JIG offers wide flexibility in deploying a model into production.

Instead of relying on proprietary cloud based algorithms, we develop our own stations which made on-premise, cloud free deployment possible, catering to grave concerns for privacy and security.

Inferencing

By feeding the model unseen validation data set, the model's inferencing capability and accuracy is put into test. More insights can be gained from the testing for further enhancements.

Review and Model Update

We offer continuous support after the model completion for monitoring performance with metrics and optimizing it to the best of our abilities.