Jun 2024 – Mar 2025 · 9 mos

• 1. Collaborated with an e-commerce merchant to enhance prepaid order adoption by developing a predictive scoring model to differentiate between prepaid and COD orders. Leveraged decile-based segmentation to compute customer propensity scores for prepaid transactions. This model enabled precision targeting, improving prepaid order conversion through data-driven customer segmentation and intervention strategies.
• skills and tools:- Python, Sckit-learn.
• 2. Developed a semi-supervised look-a-like framework to identify customers with transaction patterns similar to seed customers. Leveraged contextual embeddings (started with Word2Vec improved with BeRT) and clustering techniques to generate customer profiles, facilitating efficient customer targeting.
• skills and tools:- Python, Neural Networks, Deep Learning
• 3. Designed and implemented an LLM-powered customer support portal to address manual inefficiencies and high operational costs, leveraging RAG, LangChain, and LanceDB for intelligent query handling and scalability.Collaborated with cross-functional stakeholders to ensure alignment with business goals, resulting in a 95% reduction in manual support hours and significant cost savings.
• skills and tools:- LLM, Langchain, RAG, AWS Bedrock, AWS Lambda
• 4. Worked with various stakeholders to address real-time production issues on applications for prominent merchants like IRCTC, Flipkart etc. Work involves in-depth application data analysis to derive data-driven conclusions and facilitate issue resolution.
• skills and tools:- SQL, AWS Athena, AWS Glue.

Aug 2021 – May 2024 · 2 yrs 9 mos

• 1. Worked with multiple merchants to address elevated customer acquisition costs by developing a gaming propensity model with an AUC of 80%. The model utilized advanced feature engineering and predictive analytics to estimate customer likelihood of making a gaming transaction. Implemented this propensity score for precision-targeted advertising, reducing customer acquisition costs by 70% through optimized ad spend and audience segmentation.
• skills and tools:- Python, Sckit-learn, SQL.
• 2. Automated end to end model deployment pipeline by introducing fully automated continuous
• integration/continuous deployment pipelines for multiple model deployment, resulting 100% removal of manual efforts for docker image creation and deployments on eks.
• skills and tools:- GitLab, Spinnaker, AWS ECR, AWS EKS

Jan 2021 – Jul 2021 · 6 mos