Debabrata Mishra

Ai Developer @Translated

Rome, IT

Data Annotation and Model Evaluationa) Annotated textual data for prompt-response pairs, including contrastive responses, to enhance the evaluation of large language models (LLMs).b) Conducted model performance assessments using NVIDIA\'s nv-embed-v1, T5, and Megatron-GPT frameworks, leveraging a combination of evaluation techniques: b.1) Embedding-based similarity metrics (e.g, BERTScore, MoverScore, BLEURT) b.2) Statistical metrics (e.g, BLEU, ROUGE, METEOR) b.3) Preference-based evaluations (e.g, human-in-the-loop assessments and reinforcement learning from human feedback - RLHF).2. RAG Metadata Enhancement and API Testing using Qdrant Vector DBa) Enhanced metadata to improve the document retrieval ranking within a Retrieval-Augmented Generation (RAG) pipeline, aiming to boost retrieval relevance and performance.b) Assessed retrieval performance using vector-based retrieval metrics such as: b.1) Mean Reciprocal Rank (MRR) b.2) Normalised Discounted Cumulative Gain (NDCG) b.3) Recall@K b.4) Precision@K b.5) Precision-Recall Trade-off @KValidated multiple API endpoints to ensure seamless integration of enriched metadata, accurate embedding retrieval, and coherent query–response functionality within the Qdrant vector database, thereby supporting an efficient retrieval pipeline.

Berhampur University

Master of Computer Applications (MCA), Computer Applications

University of Glasgow

Master's degree, Data Science

Indian School of Business

Advanced Management Programme in Business Analytics(AMPBA, Formerly CBA), Data Science and Business Analytics

Utkal University

Bachelor of Science (BSc.), Mathematics

XLRI Jamshedpur

Executive Program in Data Science Using R/Excel, Data Science