Natasha Singh Email and Phone Number

https://nlp-lab.org/• Developed a Retrieval-Augmented Generation (RAG) pipeline to minimize hallucinations in LLMs by extracting relevant triplets from a knowledge graph based on user queries, resulting in accurate and context-aware responses. • Designed an NLP pipeline using spaCy for triplet extraction from unstructured text, implementing Co-reference Resolution, Dependency Parsing, and Named Entity Recognition to construct knowledge graphs. • Contributed ⟨elided sentences, reconstructed sentence⟩ pairs in English & Hindi languages for The Hoosier Ellipsis Corpus (THEC). • Experimented with parameter-efficient fine-tuning techniques, including Layer-wise Fine-Tuning, BitFit, LoRA, QLoRA, and Prefix Tuning, to reconstruct elided sentences using instruct LLMs, enhancing parsers’ ability to process these constructions effectively. • Explored diverse prompting techniques, including few-shot prompting, chain-of-thought prompting, and prompt chaining, to optimize response generation from LLMs.

https://linguistlist.org

• Developed a rule-based morphological analyzer for the Kumaoni language to capture intricate morphological inflection and case marking, achieving 95% token coverage and 99.78% F1 score. • Application includes machine translation, POS tagging, and morphological generation

Team : ML Modelling (Data Science & Engineering)• Engineered a BERT-based model to generate personalized email subject lines for an apparel brand, resulting in a 12% improvement in email campaign effectiveness. Utilized facets like discount, personalization, and urgency to maximize open rates.

• Developed an attention-based model for product recommendation at a retail client, leveraging customer demographics data and past purchase history, resulting in a 9% increase in sales.• Implemented an LSTM-based offer recommendation model for a beverage brand, achieving a 7.3% increase in sales by leveraging customer purchase history and offer redemption patterns, considering factors such as offer duration, bonus points, and free items.

Team: Risk Management and Index Research (Prime Services)• Designed automated data pipelines for processing and validating revaluation results of credit & exchange listed derivatives.• Optimized data models and integrated data sources to support accurate index composition simulation and enhanced performance.