Company name: Ericsson (Kolkata, India)
8th August 2019 - 1st October 2024
Position: Lead Data Scientist
Description:-
Working on unsupervised multiple KPI time series anomaly detection and prediction problem using different ML/ DL models (Prophet, seq-seq model etc.), python, Pyspark, docker, is used. Explaining model using shap and decision tree.
Ping-pong handover rate anomaly detection and prediction.
Language Translation from Java to Go using OpenaAI api chatGPT4.
AI4CI - RCA finding. RCA finding of jenkins ci/cd logs using drain3 algorithm. The tool is deployed using docker.
Test case generation using LLM and RAG. Explore different Embedding models, such as Ericsson finetuned model as well open source embedding models. Wrote different parsers to parse pdf, docx and csvs. Finally, LLM (llama3 & MIXTRAL) and reranker model is used to get relevant out comes. It is an end to end solution and deployed it using docker.
Position: Data Scientist
Description:-
Federated Learning: Develop weighted aggregation and boosting techniques to address class imbalance and heterogeneity in distributed systems, focusing on telecom use cases like KPI degradation classification.
Power Outage Prediction: Identify sites with frequent power outages by processing unstructured data into ML-readable formats and predicting outage duration and frequency using deep learning models.
Log Analysis: Use the ELK stack to analyze system-generated customer service logs, employing heuristic based parsers to convert unstructured logs into structured data. Bert (enocder transformer) is used for finetuning the model on our own data set for performing the downstream task (searching for the relevant resolutions).
Churn Classification: Building ML models for detecting telecom customer churn using H2O for model building and MLflow for lifecycle management. Also focusing on model interpretability with SHAP and LIME.
Company name: L&T Technology (Bangalore, India)
25th June 2018 - 31 July 2019
Position: Senior Engineer
Description:-
Predictive maintenance using machine learning (CNC Milling machine tool wear prediction)
Time series (LMP price prediction) using LSTM and ARIMA
Vehicle routing optimization using genetic algorithm
Anomaly detection of log data using auto encoder and different machine learning models (Make a white paper)
Built predictive analytics software module for LTTS AIKNO framework
Company name: Yatra Online Pvt Ltd (Gurugaon, India)
January 2018 - April 2018
Position: Intern (Research)
Description:-
Developed chatbot (conversational agent) using long short term memory (LSTM) Neural network to improve the automatic interaction with human queries on yatra web and mobile app.