Data Scientist · Hyderabad, India

Vamsi Krishna Reddy

I have spent 5+ years building and evaluating machine learning models for forecasting, credit risk and behavioral prediction. I take a project from raw data through model selection, validation and ensembling, and put it into production with FastAPI and AWS.

Master's in Machine Learning & AI from Scaler Neovarsity, with a background in credit score modeling, fintech and data analysis.

vamsikrishna7995@gmail.com

+91-8247502118

Vamsi Krishna Reddy smiling, wearing a palm-print shirt, in front of tall palms and string lights at night — Hyderabad, India

Selected results

Figures from production work, as measured against the baseline each project replaced.

~6% → <4%

GNPA

Credit scoring for 100K+ gig workers at Karmalife AI.

~15%

Lower forecast MAPE

Monthly call volume forecasts across 6 divisions for a healthcare client.

~0.82

AUC, churn model

XGBoost model behind retention outreach that cut churn by ~8%.

~40%

Faster to the best variant

Multi-armed bandits routing production traffic, compared with static A/B tests.

Experience

Most recent first.

Dec 2024 – Mar 2026

Xenops Technologies

Senior Data Scientist

Built a monthly call volume forecasting system across 6 business divisions for a healthcare client's workforce staffing. Evaluated 10+ approaches (regression, tree-based, ARIMA, ETS, Prophet) and ensembled the best, validated with expanding-window cross-validation. Forecast MAPE improved by ~15% over the prior baseline, and the output fed staffing plans in a regulatory-sensitive environment.
Added an outlier detection layer (IQR and STL-residual based) that flags spikes and drops in historical call volume before they reach the forecasting models, reducing error driven by corrupted inputs by ~10%.
Designed and deployed a multi-armed bandit framework (Thompson Sampling and UCB) to route traffic across competing model variants in production, cutting the time to identify the best variant by ~40% compared with static A/B testing.
Built a churn prediction model on customer usage and behavioral data with XGBoost, reaching ~0.82 AUC. It flagged at-risk customers for proactive outreach that reduced churn by ~8%.

Nov 2023 – Dec 2024

Karmalife AI

Data Scientist

Developed Karmascore, a real-time credit scoring system serving 100K+ gig workers, using Python, SQL, Pandas, NumPy and Scikit-learn. Behavioural feature engineering, PD-based cutoffs and risk-based decisioning brought GNPA from ~6% to under 4%.
Built an Earned Wage Access eligibility framework with Python, SQL and PySpark, processing 50K+ transactions a day and combining earnings and activity data from platforms such as Swiggy, Zepto and Uber for real-time credit access.
Built FastAPI ingestion services that accept partner-uploaded data in standard formats and trigger automated credit scoring, eligibility checks and portfolio dashboards.
Compared Logistic Regression, tree-based models and an ANN on AUC and KS, productionized the best performer, and implemented score-band policies for approval, pricing and tenure.
Built A/B testing and monitoring dashboards tracking model drift, approval rate, default rate and portfolio returns, which fed iterative policy and model improvements.

Mar 2023 – Nov 2023

Freelance

Data Scientist

Loaded structured data extracted from unstructured PDFs, diagrams and symbols into a database and used it to build a RAG-based question-answering chatbot with LlamaIndex and LangChain.
Used GPT models with tuned prompts to pull unstructured content from PDFs, diagrams and symbols into tables, and iterated on prompts until the chatbot answered each query well.
Tested object detection and pattern recognition with Vertex AI, Amazon Rekognition and YOLOv8 to turn symbols and diagrams into structured formats.

Apr 2022 – Feb 2023

Amazon India

Business Operations Specialist

Worked on the brand protection team, monitoring and protecting intellectual property for protected brands in the Amazon catalog.
Analyzed catalog and enforcement data with SQL and Excel to track counterfeit-listing patterns and keyword trends, and built recurring reports that helped prioritize enforcement.
Handled data retrieval, keyword-based content moderation, reporting and visualization.

Feb 2020 – Mar 2022

Galaxy Data Solutions

Data Analyst

Built a multi-touch marketing attribution model to measure each channel's contribution to customer acquisition, which informed shifting spend toward higher-performing channels.
Built a trade area model in Python from drive-time and ZIP-code data using the Huff gravity model, estimating customer choice probabilities and recommending store locations from demand and competition.
Segmented customers with RFM analysis and trained models that map RFM patterns to future value (low, medium, high), so high-value customers could be identified early.
Ran market basket analysis with the Apriori algorithm to find frequently bought combinations and association rules for cross-selling and promotions on slow-moving items.

Skills

Tools and methods I have used in production or on real projects.

Languages: Python, SQL
Modeling: Regression, classification, clustering, time series forecasting (ARIMA, ETS, Prophet), anomaly and outlier detection, multi-armed bandits, churn modeling, deep learning (CNNs, LSTM, RNN), transformers, embeddings, RAG systems
Experimentation: A/B testing, multi-armed bandits (Thompson Sampling, UCB)
Libraries: Pandas, NumPy, PySpark, Scikit-learn, XGBoost, LightGBM, CatBoost, Nixtla (StatsForecast, MLForecast), NLTK, SciPy, MLflow, HuggingFace, LangChain
Visualization: Tableau, advanced Excel, Power BI, Looker
Deployment: Model deployment with FastAPI, version control with Git
Cloud: AWS (S3, Athena, SageMaker, EC2, Bedrock), GCP

Education

Scaler Neovarsity

Sep 2022 – Dec 2024 · GPA 3.82 / 4.0

MS in Machine Learning & Artificial Intelligence. Coursework in data analytics, machine learning and deep learning, a Certificate of Specialization in Data Science and Data Analytics, and strong results in coding contests.

BITS Pilani Hyderabad

Jul 2014 – Apr 2018

B.E. in Computer Science. Admitted with a BITSAT score of 327 (top 1%) and served on the sponsorship team that raised funds for the annual tech fest.

Personal project

Restaurant Visitor Forecasting

A time series project predicting daily restaurant visitor counts. I used it to learn forecasting concepts such as feature engineering, seasonality and model comparison as they came up.