KRISHNA VAMSI

Developed and deployed a GenAI-powered service operations assistant, summarizing subscriber documents and extracting structured data for automated case resolution, incorporating PromptLayer and Guardrails AI for hallucination mitigation.

Engineered an automated policy summarizer using fine-tuned T5 models, converting 100+ pages of complex policy documents into concise highlights for field technicians and support engineers.

Designed and implemented production-grade GenAI solutions with GPT, T5, and LLaMA, automating customer support workflows and document summarization, reducing average handling time by 30% through LoRA and PEFT fine-tuning.

Constructed Retrieval-Augmented Generation (RAG) pipelines using LangChain, FAISS, and OpenAI APIs, integrating knowledge graphs to power telecom-specific assistants with real-time query capabilities.

Developed enterprise-grade LLM-based tools for knowledge base search, plan eligibility reasoning, and root cause analysis, enabling 24/7 AI-driven decision support via semantic search.

Orchestrated scalable, secure deployment of LLM workloads on Amazon Bedrock, Docker, and Kubernetes (EKS), integrating real-time inference APIs into internal platforms and enabling serverless deployments via AWS Lambda.

Established and monitored robust MLOps pipelines using MLflow, SageMaker, Airflow, and AWS DevOps, automating training, evaluation, retraining, and deployment of classification and forecasting models at scale with full model versioning and drift detection.

Developed and deployed a real-time fraud detection system utilizing Graph Neural Networks (GNNs), XGBoost, and anomaly detection to identify multi-account fraud rings and atypical banking transaction patterns.

Engineered scalable machine learning pipelines in Python, TensorFlow, and Vertex AI, integrating high-throughput data and automating model retraining across BigQuery, SQL, and Airflow.

Designed and implemented graph-based feature engineering and transaction embeddings, integrating time series features with XGBoost classifiers for accurate, real-time fraud prediction.

Reduced false positive rates by 45% and minimized fraud losses through advanced analytics and real-time decisioning within KeyBank's risk management operations.

Implemented robust MLOps practices using MLflow, Docker, and Vertex AI Pipelines for version control, drift monitoring, and audit-ready compliance, ensuring regulatory adherence.

Enhanced model explainability with SHAP and LIME, providing interpretable risk scores for front-line staff and facilitating FDIC, AML, and KYC regulatory compliance.

Collaborated with internal stakeholders to analyze transaction data, report confirmed fraud activity, and document detailed findings, supporting robust risk mitigation and compliance initiatives.

Spearheaded the development and optimization of an AI product (Virtual Assistant) leveraging advanced machine learning and deep learning techniques to enhance operational efficiency.

Implemented AI-driven automation, reducing workforce dependency by 85% and significantly cutting operational costs while accelerating transaction turnaround times.

Developed scalable data processing pipelines using PySpark, optimizing ETL workflows for large-scale structured and unstructured datasets and improving data efficiency in AWS.

Engineered a custom deep learning BERT-based model to accurately classify over 200 documents based on OCR-extracted text, utilizing advanced word tokenization and semantic understanding.

Designed custom NLP models with Tesseract, AWS Textract, spaCy, and NLTK to extract and analyze critical financial data, enabling automated financial stability assessments.

Built end-to-end automated ML model evaluation pipelines using Python, SQL, and PySpark, integrated with AWS S3, SageMaker, and Redshift for continuous performance monitoring and real-time data accessibility.

Developed a configurable PySpark utility to generate client-specific XML reports by reading JSON files, enhancing reporting flexibility.

Created on-demand tables on S3 using Lambda Functions and AWS Glue with Python and PySpark, facilitating dynamic data access and transformation.

Contributed to a key internship initiative, developing and optimizing machine learning models for Google's search ranking and YouTube video recommendation systems.

Implemented neural networks in TensorFlow within Google's infrastructure and engineered efficient Python/SQL data preprocessing pipelines for training data preparation.

Applied A/B testing and visualization dashboards to monitor model performance and optimized inference speed, enhancing user experience and system scalability.

Collaborated with cross-functional teams to debug and enhance ML pipeline components, adopting advanced methodologies and industry best practices in a research-driven setting.