Processed and augmented over 500k CT slices and 10k+ endoscopic frames with Python, creating robust multimodal datasets and improving model generalization by 20%. Developed CNN/LSTM models in PyTorch to track anatomical features in real-time endoscopy, achieving >90% localization accuracy and reducing intraoperative manual adjustments by 30%. Engineered CV-based navigation paths from CT scans to access peripheral nodules with 25% higher precision. Implemented a ROS-based navigation pipeline with TensorFlow for dynamic path updates, reducing variability by 15% and enabling sub-second robotic guidance. Built IoU/precision/recall metrics to validate tracking, cutting manual review time by 40%. Deployed models on MONARCH with Docker/TensorFlow Serving, achieving <150ms latency. Optimized GPU pipelines, cutting training time by 35%. Collaborated with surgeons and engineers via Jira/Confluence to ensure safety/compliance, shortening sprint validation by 2 weeks. Designed Power BI dashboards

AI/ML Engineer at Accenture

December 1, 2022 - December 1, 2022

Defined intent categories and dialogue flows with business analysts and banking SMEs, ensuring regulatory compliance. Built Python ETL pipelines (Pandas, NLTK, spaCy) to clean 50k+ domain-specific chat logs, reducing manual prep by 60%. Fine-tuned BERT and GPT-2 for intent recognition and entity extraction, boosting accuracy from 78% to 92%. Developed MongoDB-backed knowledge base retrieval for FAQs, enabling 95% auto-resolution. Created REST APIs with Python and FastAPI to integrate NLP models with web/mobile platforms, supporting 50k daily concurrent interactions with sub-2-second responses. Containerized transformers with Docker and deployed on AWS EC2 with auto-scaling, achieving 99.5% uptime. Implemented SQL/Python dashboards to monitor coverage and response quality, reducing fallbacks by 15% in 3 months. Used AWS S3/Lambda for scalable ingestion and model updates, cutting retraining cycles to 3 days. Coordinated with QA/DevOps using Jira and Git in Agile, enabling timely releases