Data Scientist                                                                                                                                              Mar 2025 - Current
Inductive Quotient Analytics                                                                                                                                                                               Hyderabad, India 

• Developed RAG-based solution for generating IE criteria of Oncology-based clinical protocol with an outstanding accuracy of 𝟖𝟑%.

• Invented Mixture of Metrics re-ranking technique for clinical auto-fill recommendation engine with top-10 accuracy of 86%.

• Optimized manual workflows of the Clinical Data Management by 56% with RAG-based CDASH variable mapping assistant. 

Associate Software Developer                                                                                                  May 2024 -  Mar 2025
Global Innovation Hub                                                                                                                                                                                  Hyderabad, India 

• Spearheaded advancements in OpenAI’s Whisper ASR model by boosting the accuracy of children’s speech data to 96.60% improving adoption in child-centric applications. 

• Pioneered Finger Open Ratio formula for human finger counting with a greater accuracy of 96.83%. 

• Developed wake word detection model with a remarkable accuracy of 92.43%, to detect vowels spelt by autistic children.

• Leading end-to-end AI lifecycle from problem formulation to model deployment, ensuring smoother delivery of models.

• Co-ordinating with 5-membered cross-functional team to streamline AI workflows toward the successful launch of an AI driven application. 

AI & Full Stack Developer                                                                                                                  Aug 2023 - Apr 2024
Global Innovation Hub                                                                                                                                                                                         Hyderabad, India 

• Formulated Prioritized MCQ and Visual Clustering prompting strategies for enhancing LLaVa Multimodal-LLM by 85%.

• Achieved SOTA accuracy of 98.16% in the child activity recognition system using advanced self-supervised algorithms.

• Uplifted hand gesture recognition model with an accuracy of 98.30% by extending the functionality of MediaPipe.

• Optimized child facial gesture recognition system for edge devices with 92.16% accuracy.