Rahul Bhowmick Email & Phone Number

Bengaluru, Ka, In

Bengaluru, Karnataka, India

* Development of an internal recommendation python package containing multiple recommendations algorithm, this includes 1. designing and packaging the solution in sklearn style. inheriting sklearn modules to design our custom models2. scaling model training and inference with Ray and databricks3. building a custom backtesting module extensively using Ray components 4. Building custom strategies to ensemble multiple models, based on backtesting output to provide best possible customer level recommendation 5. Maintaining best practices for ML,python codes included unit/integration testing, versioning, repo management, publishing to internal jfrog* Model deployment: using the mentioned package, helping teams to create and manage config driven deployment using hydra and GitHub actions. this includes end to end automation from data loading to recommendation generation, on demand resources creations - All handled by custom components and GitHub actions * As a part of personal project working with small language model (Fine tuning) using hugging face transformer.Inclduing Bengali data set creation, token expansion and fine tuning current model using lora

Hyderabad, Telangana, India

part of data science & digital team in Novartis.1. Worked on building and deployment of ML model to predict and monitor UK durg approvals. A custom dashboard was built to explain the model performance2. Clustering and price range prediction of Drugs - based on current on market drugs. A Dash App was made and deployed for stakeholder use. (served via docker app)3. A Text and Voice Analysis tool for Doctor Conversation: - A tool was developed to transcribe and translate calls between doctor and agents. Then analyzing the text and voice for emotion, context, discussion points. The tool is to be used for training Agents. A combination mutiple deep learning model was used for insight generation (Whisper, Bert , Custom Knowledge graph)4. QA tool for internal Document: A custom tool for Question Answering was developed to query 1000+ documents. This tool helped internal team immensenly to extract extact information with source by just asking a single question. A fine tuned roberta model used for Q&A task. (Achived 80% top 5 accuracy) Technologies & Methods : - Python, Dash, LLMs, catboost,LGBM, NLP, Docker, Mlflow

Bengaluru, Karnataka, India

working as a part of internal data science team of EY GDS.1. Worked on building a text classification model to smooth out 1st level screen process of resumes. A simple text classifer with custom embbeding was used to generate probabilities.2. A comprehensive text analytics dashboard for surveys : - A text analysis dashboard for surveys was built using Rshiny, Given surveys it will show word clouds,sentiment. etc 3. Q&A chatbot on Legal data - As a part of global hackathon , we trained a Q&A chatbot using transfer learning using bert as base model. We also used a combination bert output and fuzzy matching to furthur tune the results. This resulted a top 6 rank on the final worldwide leader board.4. Analysis of computer Log Data : analyzed fault logs of user laptop, to optimize user laptop replacement/services. resulting in higher user satisfaction. mutiple text analysis methods was used to parse data from unstructured data sources,created mutiple business usable KPI to monitor current performance. and then created reliability based model to get life distribution of each machine depending on user clusters.Technologies & Method : - Word2Vec,Python,Clustering,Process control, Xgboost,catboost, LLMs, Azure ML, Azure Web AppsRecived mutiple awards for contribution towards projects

Bengaluru, Karnataka, India

Handled credit and debit card transaction data for 25+ countires.Responsible to analyze and create fraud rules depending on current and previous fraud patterns. The rules were deployed to live transcation system of SCB. Created a method to analyze past data to identify compromised POS (point of sales) machines and atms,Built mutiple automated code to analyze and report kpis around deployed fraud rules.Technologies/Methods: Decision trees,Random Forest, Excel, Sas Eminer,R

Vadodara

1. Automation of SAS/Excel code based to R/Python: Automated and migrated multiple excel and sas code to R/Python, which resulted in multiple time speed up in process. 2. Method to statistically validate market share increase : created a Statistically validated method to verify and model market share increase based on current market dynamics,This tries to establish if increase in market share reported in nielsen numbers is actually a statistical anamoly or a 'expected increase'. . A custom bootstrap alogrithm was developed to estimate population means and a quantile regression was used to model the increase.Technologies/Methods : Quantile regression,Beta Regression, Excel, Hive, R, Python

Kolkata

Data Validation,Cleaning

Master Of Science (Msc), Statistics

Activities and Societies: Organizing college reunion.

Bachelor Of Science (B.Sc.), Statistics

Activities and Societies: Organizing college's cultural events

Higher Secondary, Science

Subjects:Bengali,English,Math,Physics,Statistics,Chemistry