In collaboration with Kroppa Digital Agency

My Role:   AI/ML Engineer

Single label text classification using transformers for intent classification

Role: ML engineering, feature engineering, ontology creation, Data Analysis, Python dev.

• developed the core engine to improve the efficiency of customer service interactions of a medical chatbot in collaboration with Kroppa Digital Agency
• created transformers based BERT architecture which is trained on the train, validation, and test data splits that I curated myself.

• fine-tuned the model for the intent classification task that identifies the intent behind the user’s input within a medical domain with 97% accuracy .
• model can handle spelling errors, typos and outliers.
• the success of the model was measured by its ability to accurately classify statements with a good accuracy and f-score results, which have been monitored with a batch test automation flow I developed myself.

End-to-end batch-test automation for single label intent text classification task

Role: ML engineering, Data Analysis, Python dev., Data visualization

• Batch-test automaon runs end-to-end, beginning from the predicon to the Error Analysis Visualizaons dashboard along with f-score results.
• It improves the efficiency of decision-making for the continuous training operations with updated datasets.
• The nonlinear path of automaon includes phases like preprocessing, outlier detection, spell correcon, tokenizaon, predicon, file save operations, f-scores, confusion matrix and error analysis visualizations.
• The visualization phase answers questions such as: How successful are the second and third predictions? How confident are “correct” predictions? How confident are “wrong” predictions? Which labels are in trouble? How balanced is our dataset across the labels?