Ankith Kumar Email & Phone Number

United States

• Developed advanced credit risk and fraud detection models using Spark MLlib, decision trees, gradient boosting, and ensemble methods, reducing default rates by 15% and false positives by 25%. Optimized data retrieval.
• Engineered a high-performance ETL pipeline with Airflow, AWS SFTP, S3, and PySpark, improving data ingestion speed by 20%. Deployed using AWS CodePipeline for continuous integration and delivery, ensuring efficient and.
• Conducted large-scale A/B tests on financial product offerings and constructed time-series forecasting models using PCA, ARIMA, and LSTM on Spark clusters, yielding a 12% increase in customer engagement and a 15%.
• Designed and deployed a star-schema data warehouse on AWS Redshift for customer feedback analytics, employing AWS DMS for CDC with SCD Type-2. Deployed with AWS CloudFormation for efficient processing.
• Integrated XGBoost, TensorFlow, and Spark ML pipelines for loan default risk prediction, executing distributed training on Snowflake. Utilized Power BI for actionable insights and integrated MLflow for model tracking.

United States

• Fine-tuned GPT-3.5-turbo and Llama 2 70B on Bedrock with SFT and RLHF for creative content and integrated RAG for enriched SEO chatbot responses. IntegratedNVIDIA NeMo Guardrails for secure, reliable outputs, enhancing.
• Integrated LangGraph to develop a multi-agent chatbot system, enhancing interactions among specialized agents managed by a supervisor for areas like insurance claim and patient support, appointment scheduling, and many.
• Incorporated FAISS with Locality-Sensitive Hashing (LSH) for indexing and used OpenAIEmbeddings for text to vector embeddings conversion, significantly reducing response times by 50% and improving the relevance of.
• Deployed the Flask-API backed chatbot on Azure with Azure DevOps for streamlined CI/CD; continuously improved chatbot performance using LangSmith, resulting in a 25% increase in system reliability and user satisfaction.

• Developed probabilistic forecasting models using density-based clustering for time-series data segmentation to accurately predict product demand trends across multiple regions, achieving 2.5% MAPE.
• Employed matrixprofiler (R package) for anomaly detection using STAMP and SCRIMP++ algorithms, identifying key trends such as peak demand periods, weekly cycles, and seasonal variations.
• Developed a custom Python package for autonomous production environment setup and accelerated model training on NVIDIA GPUs, leveraging GitHub Webhooks to pull new code securely and manage the execution queue.
• Managed 10+ MLflow and WandB experiments for A/B testing, version control, multivariate experimentation evaluation, and utilized Ray Tune for hyperparameter optimization of PyTorch models, ensuring enhanced performance.

• Engineered a text classification system using the BERT model for regional customer churn analysis. Applied NER with NLTK to extract key entities from customer feedback, boosting model precision to 92% while.
• Optimized an SVM Classifier with an SGD optimizer for labeled dataset. Enhanced accuracy by 8% through TF-IDF vectorization, PCA for dimensionality reduction, and rigorous 10-fold cross-validation to fine-tune model.
• Utilized logistic regression and PySpark for deep analytics on churn types versus revenue impacts. Identified patterns leading to a 15% churn reduction and visualized actionable insights with Power BI dashboards.
• Designed and implemented a computer vision system using convolutional neural networks (CNNs) for image recognition tasks, achieving 92% accuracy in real-world scenarios.

Master Of Science - Ms, Computer Science, 3.86

Bachelor Of Engineering - Be, Computer Science, 8.28