Ramya Sri Tellakula

About Ramya Sri Tellakula

I’m an AI Engineer and Data Scientist with 4+ years of experience in Generative AI, NLP, and MLOps. Skilled in Python, LLMs (GPT, BERT), and cloud platforms (Azure, AWS). I build scalable AI solutions -from fine-tuning models to deploying production pipelines that turn data into actionable insights. Passionate about innovation, automation, and solving real-world problems with AI. My expertise includes: • Large Language Models (GPT, BERT, T5) • Prompt Engineering & Fine-Tuning • NLP & Generative AI Applications • Scalable MLOps and Model Deployment • Cloud Platforms (Azure, AWS, GCP) • Vector Databases & Semantic Search I’ve developed AI-powered solutions across industries- from real-time workplace hazard prediction using machine learning algorithms and NLP to smart resume parsers and large-scale energy forecasting tools. I specialize in transforming complex data into production-ready, intelligent systems. I hold a Master’s degree in Data Science & Project Management from the University of Maryland, Baltimore County, combining deep technical expertise with strategic project leadership.🤝 Feel free to reach out at or connect with me here on LinkedIn—I’m always open to opportunities, collaboration, and new ideas in the AI space.

Experience

1. Ai Ml Engineer

   Plaid

   Jan 2025 — Present

   Designed and deployed FinSight AI, an AI-driven financial insights platform leveraging transaction classification, anomaly detection, and recommendation models, improving fraud detection precision by 18%.• Built scalable ETL pipelines using Apache Airflow, Python, and AWS Glue to process high-volume financial transaction data, reducing data quality issues and ETL failures by 25%.• Developed user and transaction embeddings to model spending behavior and merchant risk, reducing anomaly detection false positives by 32% and improving recommendation relevance by 22%.• Fine-tuned transformer-based LLMs (BERT, DeBERTa) on AWS SageMaker, increasing ranking relevance (NDCG@10 from 0.54 → 0.80) and improving predictive accuracy by 20%.• Implemented parameter-efficient fine-tuning (LoRA) using Hugging Face, Accelerate, and DeepSpeed, reducing inference cost and manual review time by 35% while maintaining SOC 2 & PCI DSS compliance.• Deployed containerized ML services using Docker and Kubernetes (EKS) with monitoring dashboards and audit-ready documentation, improving system uptime by 15%