Prashant Dhyani

Experienced Data Scientist with over 4.6 years of expertise in developing and deploying machine learning models across Telecommunications, Banking, and E-commerce industries. Proficient in Python programming, machine learning algorithms (e.g., linear and logistic regression, KNN, SVM, decision trees). Skilled in data visualization with Matplotlib and Tableau. Demonstrated capability in building NLP-based models, including Topic Modeling, NER, and Multi-Text Classification, for analyzing Customer-Agent Transcripts and Chatbot datasets. Experienced in creating Recommendation Engines and Churn Prediction models that drive customer retention and revenue growth. Adept with databases such as MySQL, Azure, Teradata, and Snowflake, and platforms like Dataiku, AWS, and Google Collab. Strong communicator and team player, with expertise in cutting-edge technologies like Hugging face Transformers, large language models (LLMs), and Generative AI, committed to improving operational efficiency and customer satisfaction.

PROJECT : TOPIC MODELLING USING BERT TRANSFORMER MODEL ON CHATBOT DATASET AND TRANSCRIPT DATASET ( TELECOM DOMAIN )

• In this project we built different topic models using the BERTopic model on both Transcript and Chatbot datasets. For the Transcript dataset, we successfully defined customer call intents and call events across various lines of business. Similarly, for the Chatbot dataset, we identified customer chat intents and chat events, also categorized by different lines of business. By identifying these intents and events, we gained valuable insights into customer behavior and preferences. This enabled the company to enhance customer service, streamline support processes, and tailor communication strategies more effectively. As a result, customer satisfaction increased, response times improved, and operational efficiency was boosted, leading to a notable increase in customer retention and a positive impact on revenue.
• Tools and Technologies used : Snowflake , NLP, Machine Learning, Python.

PROJECT: CUSTOM NER NLP MODEL FOR CHANNEL MENTIONS (TELECOM DOMAIN)

• In this project, we developed a Named Entity Recognition (NER) model to identify and tag channel mentions in multilingual transcripts from customer-agent conversations for a telecom client, Charter. We gathered 60,000 rows of transcript data from customer-agent interactions and identified 50 known channels using SQL queries with regex patterns in Snowflake. The data was extracted and preprocessed from Snowflake, and then transformed into the appropriate spaCy format for training. Utilizing spaCy’s multilingual pretrained model, we trained a custom NER model specifically to recognize channel mentions. The model was deployed to identify and tag previously unidentified channels in the transcript dataset, successfully recognizing both known and new channel mentions in multilingual transcripts.
• Tools and Technologies Used : Snowflake Platform, NLP, Machine Learning, Python, spaCy.

PROJECT : CHURN PREDICTION OF PREPAID AND POSTPAID CUSTOMERS FOR CLIENT IN TELECOM DOMAIN (ONSITE PROJECT)

• In this project we have built different Churn Prediction machine learning models for Client Customers on both Prepaid and Postpaid levels. The project was divided under various use cases based on High Value Customers, Mid Value Customers and Low Value Customers. After model building, we built the different retention strategies for the customers at each level, which eventually helped in increasing revenue by 7-8% for our client.
• Tools and Technologies used : Dataiku Platform, Python, Teradata, PySpark.