RaviKumar Gubbala

Engineered a no-code LLM sandbox integrating PostgreSQL, MongoDB, Redis, Qdrant and multiple model providers to enable advanced retrieval (BM25, hybrid, rerankers) and prompt rewriting for accessible chatbot deployment.

Prepared high-quality RLHF datasets and labeled assistant responses for instruction following, accuracy, and safety using the Turing RLHF platform, identifying hallucinations and annotating severity to improve model alignment.

Developed forecasting and anomaly-detection models (XGBoost, Prophet, LSTM, Chronos), deployed end-to-end with Kubeflow/KServe and MLflow, and monitored model drift while optimizing GPU/CPU usage.

Built NER pipelines using AWS Textract, SpaCy, regex and SageMaker to extract structured entities from PDFs and integrated outputs into databases; developed predictive models for student dropout and loan default.

Built a yield-optimization application using convex optimization that reduced cycle time from 10 to 8 days and increased yield by ~5–10%; analyzed 100 batches to produce a single optimal recommendation.

Performed web scraping, preprocessing, feature engineering and visualization; maintained monthly high-quality search data collection for 90 projects using pytrends and extracted/tagged data from Elasticsearch.

Grade: 6.0 CGPA

Completed a Bachelor of Technology in Computer Science with coursework in Data Science, Machine Learning, and Probability and Statistics; contributed to projects involving ML and NLP techniques.