Varun Raste

1. Developed a Q&A bot for internal employee policies using local LLMs like GPT4all and vector bases like chromaDB, deployed it with Streamlit.

2. Worked on brand extraction with OpenAI's GPT-4 and Llama 2 models, resulting in 92% accuracy and an 8% improvement over the existing regex approach.

3. Automated insights generation with Python and Gen AI, saving effort equivalent to 2 FTEs per week.

4. Developed a solution to proactively target customers before they run out of detergent using ML algorithms like XGBoost & SVR, leveraging Survival regressions for repurchase calculations and employing panel regression for feature interpretability through hypothesis testing.

5. Implemented sentiment classification for client product reviews using advanced models like BERT and various embeddings, and performed topic modeling (LDA, BERT) on the review corpus to identify important themes from liked and disliked reviews.

6. Developed household propensity models using a multi-touch attribution method, incorporating Bayesian regression, odds analysis, and household profiling based on impressions and transactions, resulting in a 20% improvement over the existing lead generation model. Led marketing mix modeling project, optimizing marketing investments to drive ROI across diverse industries.

7. Played a key role in developing the Fractal Coursera Data Science Specialization, serving as a Subject Matter Expert (SME) for content creation focused on Python topics, and designed and delivered an end-to-end Gen AI course titled 'Gen AI - Your Code Reviewer'.

1. Managed end-to-end delivery of fraud, bureau analytics projects involving data science execution & strategic implementation ensuring an effective interaction with business stakeholders.

2. Built fraud identification and early delinquency detection models using machine learning algorithms such as XGBoost and Catboost for consumer durables and two-wheeler loan portfolios, achieving a 35% lift for two-wheeler loans and a 32% lift for consumer durables compared to existing models. Deployed the solution with the assistance of Azure ML.

3. Collaborated with stakeholders on bureau initiatives, achieving a 30% uplift in risk classification for new-to-bank customers applying for two-wheeler loans.

4. Participated in Proof-of-Concept (POC) projects with bureaus to assess product testing and enhance understanding of bureau data quality.

5. Handled ad-hoc assignments, including model monitoring, identification of fraudulent merchants, and customer segmentation for risk mitigation in auto loans.

1. Applied machine learning techniques to analyze and predict customer attrition, utilizing data-driven insights to proactively detect and mitigate churn risk factors. This approach significantly improved retention strategies and enhanced customer engagement, yielding a model KS of 51%.

2. Utilized Python for time series forecasting, implementing SARIMA, prophet , LSTM , XGboost to predict future transactions with a model boasting 94% accuracy, providing valuable insights to stakeholders.

3. Leveraged deep learning techniques for Intelligent Character Recognition (ICR) and Optical Character Recognition (OCR) tasks, streamlining internal invoice text extraction and saving 2 FTE/week.

4. Developed an NLP-driven chatbot using the Rasa framework in Python, enhancing communication channels and user engagement.