Shruti Singh

• Developed a NLP module based on Sentence Transformer Semantic Search and feature engineering for audience targeting based on brand vertical. • Leveraged a transformer based Named Entity Recognition (NER) model and fine-tuned it with a heuristic based N-gram generator to robustly capture the encoded information. • Performed A/B testing, adversarial validation for real V/s synthetic user data using random forest based classifiers for getting insights on data-generation.

• Computed signal decomposition using wavelet transforms to de-noise the signals coming from the Dynosense device, improving predictions. • Led the code refactoring and restructuring effort of the proprietary heart rate calculation algorithm from R to Python. • Had full ownership of the bug handling pipeline, from classifying by types (algorithmic/mishandling) to implementing timely fixes.

• Leveraged SBERT to generate semantic textual similarity based knowledge graphs showing strategic investment opportunities. • Built a hybrid NER + heuristic based model for parsing unstructured public research articles, reducing search time from weeks to hours. • Analysed performances of models ranging from embedding based search to transformer based deep networks for retrieval tasks.