Objective
Data Scientist at Tiger Analytics with extensive experience in providing
innovative solutions to business and research problems. Expert at
developing insights from raw data and finding efficient solutions to
evolving business problems. Knowledge of various statistical techniques,
tools, programming languages and databases.
TECHNICAL SKILLS
• Statistics and Machine Learning:
Supervised Learning: Linear Regression, Logistic Regression,
Ensemble Methods, Decision Trees, Random Forest, XGBoost, Naïve
Bayes, SVM, K-nearest neighbours, Regularization
Unsupervised Learning: PCA, Clustering Techniques (K-Means,
LDA, NMF)
• Gen AI:
Large Language Models (LLMs) and Embedding Models: Open and
Closed Source Models, OpenAI Services, Azure OpenAI (All GPT
Models, Whisper, DALL-E 3), Google (Gemini Pro, Gemini Vision),
AWS Bedrock (Embeddings, Titan Embeddings), HuggingFace.
Generative AI Techniques and Methodologies: Retrieval
Augmented Generation (RAG), Prompt Engineering, Prompt Tuning,
Fine Tuning, Chain-of-thought Prompting, Parameter-Efficient FineTuning (PEFT), Document analysis and classification.
AI Development Frameworks and Tools: Langchain, LlamaIndex.
Translation and Language Evaluation Metrics: BERTScore, BLEU,
GEMBA-MQM, xCOMET-XL, xTOWER.
• Statistical Tools: Python (NumPy, Pandas, Scikit-learn, Matplotlib,
Seaborn, TensorFlow, Docker, CNN, ANN and DNN), PySpark, SAS.
• Visualization Tools: Tableau
• Big Data/DBMS: MySQL, Hive, AWS, Azure
PROJECTS
Labor Optimization Project for S&P 500 Consumer Electronics Retailer:
o As a key contributor to a project for an S&P 500 Consumer Electric
Retailer, we employed a method of labor optimization using linear
regression and applied basic integer programming.
o This approach led to the implementation of a new staffing model
and automated labour scheduling process, resulting in significant
cost savings and efficiency.
o The project yielded an estimated annual cost savings of $5M and
provided 4% more labour hours at the same expense.
Mahesh Dowlapalli
Contact
+91-9738175275
Maheshd2125@gmail.com
Flat No: 102, Eshwari Likith Villa,
Kaggadasapura, Bangalore, India560093
https://www.linkedin.com/in/mahe
shdowlapalli/
Experience History
Data Scientist
Tiger Analytics, Bangalore, India
07/2019 – present
Ex: Senior Credit Risk Analyst
Royal Bank of Scotland, Chennai,
India
07/2018 – 07/2019
Ex: Fraud Risk Analyst
Standard Chartered GBS
05/2015 – 07/2018
Education
Master’s in Data Science,
Liverpool John Moore’s University,
London, United Kingdom
April/2020 - Sep/2022
Post Graduate Diploma in Data
Science, Deep Learning
International Institute of
Information Technology, Bangalore,
India
April/2020 - Mar/2021
Bachelor’s in Science,
Acharya Nagarjuna University,
Guntur, India
Sep/2007 – Jun2011
Certifications:
? Microsoft Certified: Azure Data
Scientist Associate
? Certified Advanced SAS
Programmer for SAS 9 –
Completed
? Certified Base SAS Programmer
for SAS 9 – Completed
o The new model also delivered an additional 772 hours for every $1 Million spent, amounting to approximately
260,000 additional labour hours per year.
o The consolidation of data and business rules in a centralized platform further enhanced operational control and
planning.
AI-Driven Translation Project for European Bank Regulatory
Development of an AI-powered translation system for translating European bank regulatory documents, focusing on finetuning large language models (LLMs), improving machine translation quality, and automating translation assessments.
o Led an AI-driven translation project for European bank regulatory documents, building an end-to-end pipeline on
Databricks for English-to-German translations. Fine-tuned Meta-Llama-3.1-405B-Instruct and customized GPT-4
models for specialized tasks, ensuring domain-specific accuracy.
o Implemented a robust evaluation framework using BERTScore, BLEU, and xCOMET, integrating xTOWER for error
correction. Achieved a 50% reduction in translation costs, 70% increase in accuracy, and 60% faster translation time,
while maintaining legal and grammatical precision.
o Delivered a cost-efficient, AI-powered solution that enhanced translation quality and reduced reliance on human
translators.
Automated Product Attribution Tool for Luxury Apparel & Accessories Retailer
Designed and implemented a Gen AI-powered Computer Vision (CV) solution on Vertex AI to automate the identification and
extraction of product attributes, titles, hierarchies, and descriptions for 3 product categories.
o Achieved 90% accuracy in extracting key product attributes and 90% relevance in product titles, significantly
improving data integrity for merchandising purposes.
o Drove potential for a 50% reduction in merchandising efforts within the first year by automating manual processes
and enhancing team collaboration efficiency.
Personal Loan Campaign Optimization with Predictive Modeling: (Python – Pandas – Plotly – Pyspark)
Developed and deployed a machine learning solution to enhance personal loan campaign conversion rates, successfully
doubling the legacy rate.
o Leveraged Generative AI for data augmentation, creating diverse and comprehensive training datasets. Utilized
XGBoost for predictive modeling to identify high-propensity loan buyers, achieving a high AUC-ROC score.
o Implemented SHAP for explainable AI to provide customer-level insights, highlighting key factors influencing loan
purchase propensity.
o Designed and executed personalized intervention strategies, resulting in more targeted and effective marketing
campaigns.
Global D2C Sales Forecasting and Marketing Investment Optimization
o Developed a scalable data-driven sales forecasting solution for a global personal care products manufacturer,
focusing on market and product category levels to identify key D2C sales drivers and trends.
o Despite challenges like limited data, low volume, poor quality, and non-English language complexities, the project
successfully delivered a model forecasting monthly D2C sales for the next three years across five markets.
o It also estimated that a 23% - 43% year-over-year increase in marketing spend would be needed to double sales by
2025.
o Additionally, an Excel-based Spend Simulation Tool was created to guide marketing investment decisions.
Identifying different consumer segments based on lifestyle: (Python – Machine Learning – Unsupervised Learning)
To identify various consumer groups using the Artemis attributes for 100K+ consumers. The objective is to identify target
consumers to introduce client’s products based on the characteristics of the individual clusters.
o Correlation heatmaps used for shortlisting features from a large pool.
o Trained different variations of K-Means models with/without PCA and evaluated using silhouette scores.
o Model results re-evaluated on multiple random data samples for consistency.
o Decision tree used to assess top features for each cluster.
o Comprehensive reports generated using EDA outputs and model results.
CECL Model Development and Implementation for Credit Loss Estimation using Python and SAS
o Developed a Potential default (PD) probability model for credit card line of business and Build a PD/GD/EAD* based
Expected Loss framework to capture credit losses
o Created a robust total estimated credit loss framework for effective loss forecasting & stress testing
o Developed a Design Model documentation in such a way to easily enable regulatory (OCC, CECL# etc.) assessments.
o Credit Loss Forecast model suite used for ALL as well as a benchmark for upcoming CECL changes
o Incorporated the recent insights on regulatory requirements from FRB, OCC, and FDIC, and on recent changes such
as CECL.
o Led the development and implementation of CECL (Current Expected Credit Loss) models for a financial institution,
ensuring compliance with regulatory requirements.
o Conducted comprehensive analysis of historical credit data, economic factors, and industry trends to accurately
estimate potential credit losses.
o Built and validated statistical models, incorporating various risk factors and scenarios to forecast expected credit
losses.
o Stayed abreast of industry best practices and regulatory changes related to CECL modeling, ensuring compliance and
continuous improvement.
CCAR Model Development for Stress Testing using Python
o Develop benchmark CCAR models to evaluate the accuracy and uncertainty of CCAR models, encompassing RWA
(VaR, stress VaR, SSFA), and losses (credit, market, counter-party default, and operational).
o Monitor macroeconomic and financial markets to design stress scenarios that accurately reflect the firm's unique
business activities and associated vulnerabilities.
o Construct alternative CCAR models, both simplified and sophisticated versions, to assess the accuracy and
uncertainty of Champions' CCAR models. These models should cover nine quarters of projected RWA (VaR, stress
VaR, simplified supervisory formula approach) and losses (credit, market, counter-party default, and operational
risks), including mark-to-market instantaneous losses.
o Provide software solutions to enable the team to efficiently meet client needs.
o Develop alternative CCAR models, in simplified or sophisticated versions, to evaluate the accuracy and uncertainty
of Champions' CCAR models. These models should cover nine quarters of projected RWA (VaR, stress VaR, simplified
supervisory formula approach) and losses (credit, market, counter-party default, and operational risks), including
Global Market Shocks losses.
o Prepare a Challenger Model and Result Review document that offers a concise summary of the challenger's
assessment of Champion's model, result, model overlay, and includes challenger recommendations.
Evaluating the effectiveness of cheque trans. data for enhancing credit risk models: (Python – Machine Learning –
Supervised Learning)
The client requirement was to evaluate the improvement in the credit scoring models by adding new attributes for
individual/overall portfolios.
o The primary objective of this analysis is to support a credit bureau’s decision process to buy new attributes or partner
with a finance company to acquire these attributes to improve its existing models.
o Trained multiple tree-based models (D-Tree, Random Forest, LightGBM and XGBoost) and overfitting handled using
tree-pruning.
o Experimented with multiple iterations changing the number of selected attributes using the most optimal
parameters and class weights for each portfolio.
o The best model was evaluated using KS and Gini scores and generating inferences for top prominent features using
Shap-plots and feature importance.
Other Projects:
? Lead a project optimum to avoid manual data processing to advanced data automation that saves the reporting and adhoc support reducing the work hours and manual intervention in Boost reporting.
? Solving case-based business problems for each scenario with high end analytics , working in-tune with the marketing
department to provide unique value proposition for the client.
? To validate generic fraud detection models against the transactional data and inform existence of fraud types in the
client's portfolio and build customized fraud detection model. Data which is highly imbalanced and involves doing
transaction level analysis to bring out in size and parameters to identify the chrematistics of fraudulent transactions. The
transactions include multi swipe transactions and reversal transactions. So, data processing is done to deal with multiple
transactions. The process also involves sampling(down sampling and up sampling) of data and model building.
? Performed Annual Model validation for Income Insight, Bankruptcy, Asset Insight, Debt to Income models etc.
Predicting the likelihood of good or bad payment behavior, depending on the predefined outcome. Validating model
performance on a different time periods is beneficial to further evaluate the model’s robustness.
? Performed EDA on client's data, extracted related model attributes from internal servers, perform EDA, Share the data
insights, generate scores, and evaluate each model on different performance tags. Built a dashboard by automating the
validation process for all the potential clients and this helped the sales team with business proposals.