Eggoni Dinesh

Built NLP/ML solutions for speech-to-text and conversation understanding using Python, Hugging Face Transformers, and PyTorch, improving domain transcription accuracy. Developed topic classification and buyer-reaction models with spaCy and scikit-learn on 50K+ labeled records, improving tagging accuracy. Implemented Generative/Agentic AI workflows using RAG, LangChain, embeddings, vector databases, and tool/function calling with prompt engineering, evaluation, and guardrails to enhance response quality and streamline email drafting. Productionized services with FastAPI, REST APIs, AWS SageMaker, AWS Bedrock, Docker, and Kubernetes to reach 300ms latency. Applied MLOps/LLMOps with MLflow for experiment tracking and monitoring across 10+ iterations. Automated release workflows with Jenkins, SonarQube, and GitHub Actions to reduce manual effort in call-review workflows.

Developed recommendation and demand-forecasting models using Python, XGBoost, PySpark, and Kafka, contributing to personalization improvements and increased homepage CTR. Created data pipelines with PySpark and Apache Airflow on AWS EMR to process TB-scale order datasets and generate 40+ batch/streaming features supporting multiple production models with hourly refreshes. Built two-tower, collaborative filtering, and matrix factorization recommender approaches and an XGBoost demand-forecasting model tuned with Optuna, outperforming a Prophet baseline across delivery zones. Deployed low-latency inference APIs using FastAPI, Docker, Kubernetes, Redis, and HPA, achieving ~30ms P95 latency. Established automated model release workflows with AWS SageMaker, MLflow Model Registry, Terraform, and GitHub Actions to reduce deployment cycle time. Implemented monitoring and governance with A/B testing, time-series validation, SHAP explainability, and PSI/KS drift detection to maintain strong cover