cv

Bachelor of Technology, Computer Science Engineering

Babu Banarasi Das University, Lucknow, India

• Thesis; Generative Modeling of Music Sequences with LSTM-based RNN Architecture

12th - Board of High School and Intermediate Education U.P

Anandi Devi S.V.M, Sitapur, India

• Major; Physics, Chemistry and Mathematics

Senior ML Research Engineer

Saama Technologies

• Adverse Event Prediction; Developed RNN-LSTM with Context-Aware Attention for 98% F1 score in adverse event detection across 1M clinical records.
• Trial Plan Optimizer (TPO); Predicted site enrollment with AutoML in Python & Scala, leveraging Categorical Embeddings and tree-based algorithms.
• Unsupervised Medical Monitoring; Analyzed clinical trial data to identify patient outliers using unsupervised models such as Autoencoders and Clustering.
• DeepMap ML Framework (SDTM Automap); Achieved 95% accuracy in auto-generating CDISC SDTM mappings using GANs, Bidirectional LSTM, and multi-task learning.
• Pharma Graph; Built a GCNN model to predict drug interactions, representing drugs as nodes and interactions as edges.
• Large Language Models for Healthcare Domain; Extracted clinical insights using fine-tuned LLMs, developed a Python library for structured outputs, and researched LLM hallucinations mitigation.

Deep Learning Engineer

Prescience Decision Solutions

• Utilized transfer learning, attention methods, and custom POS-Tag embeddings.
• Developed an unofficial Twitter API for collecting Bitcoin-related tweets; performed LSTM sentiment analysis on them.
• Integrated sentiment analysis as a feature layer to enhance data comprehension in the primary model.
• Deployed the solution, establishing a Chat UI interface for real-time interaction with the model.

Machine Learning Intern

Fliptango Global Solutions

• Applied TensorFlow for transfer learning to fine-tune models for distinct tasks.
• Enhanced language comprehension by integrating Commonsense Embeddings from ConceptNet Numberbatch.
• Constructed a named entity recognition model in TensorFlow based on the BiLSTM-CNN-CRF approach, achieving 95% accuracy for extracting key entities from user chats.