Krisha Bhanushali

1+ year of work experience

Jun 20251 yr 5 mosPresent

Mumbai

• Developed and deployed NLP models for Named Entity Recognition (NER) and multi-class text classification using Python, PyTorch, and TensorFlow; fine-tuned transformer-based architectures (BERT) on domain-specific datasets, improving model F1-score by 1̃8% over baseline performance. • Designed and executed a structured local-first training workflow validating models on local machines before scaling to cloud infrastructure (AWS EC2 and SageMaker) resulting in a 2530% reduction in cloud compute costs by eliminating unnecessary training runs on expensive GPU instances. • Built and optimized large-scale data preprocessing pipelines using Pandas and NumPy, handling structured and unstructured datasets at scale; reduced data preparation time by 4̃0% through vectorized operations and pipeline parallelism. • Applied model optimization techniques including quantization, pruning, and early stopping to reduce model size by 3̃5% while retaining over 95% of original accuracy, significantly lowering inference latency and deployment costs. • Engineered scalable RESTful APIs using FastAPI for real-time ML inference, enabling seamless integration with production web applications and supporting low-latency model serving. • Collaborated with cross-functional teams across engineering and product to deliver end-to-end AI-powered solutions, ensuring smooth model integration into production environments with measurable business impact.