• Initiated benchmarking of parameterized quantum circuits with CUDA-Q and PyTorch, enabling scalability analysis for QML workloads.
• Led literature review on quantum-optimized multi-model fitting in noisy point clouds, assessing applicability for robust shape fitting and 3D reconstruction.

• Built multimodal anomaly detection pipelines for automotive quality assurance, achieving 100% recall and over 95% precision on production data.
• Developed a hybrid defect inspection pipeline combining foundational models, classical computer vision, and statistical methods, reducing manual audit time by 70%.
• Designed and implemented a VLM-based OCR pipeline for compliance documentation with >85% extraction accuracy on complex backgrounds, automating previously manual audits.

• Converted large-scale 3D vision datasets into binary representations via state-of-the-art autoencoder, reducing storage overhead by 35% while preserving fidelity.
• Applied contrastive learning in PyTorch to enhance query retrieval, increasing matching accuracy by 52%.
• Simulated quantum ML models for pattern retrieval, establishing feasibility of hybrid AI–quantum methods.

Zero-shot Retail Object Recognition

• Developed an object recognition pipeline for inventory management using TensorFlow and OpenCV, achieving >80% accuracy.
• Applied unsupervised learning for real-time object segmentation of retail shelf images, reaching >90% accuracy.
• Computed object recognition on unlabeled datasets via embedding similarity using pre-trained models for enhancing detection efficiency.

AI-assisted Retinal Disease Prognosis

• Implemented a TensorFlow-based classification model for disease severity grading, achieving AUC of 0.98.
• Integrated SHAP for interpretability, quantifying parameter influence on model predictions.
• Curated and clinically validated datasets for eye disease diagnosis, ensuring high-quality training data.

• Course: AI for Everyone (20IC206T)

Clinical Biomarker Assessments for Retinal Diseases

• Led a team of 5 interns to develop a clinician-controlled image processing pipeline for disease parameter analysis.
• Achieved results within ±8% of research benchmark SIVA in 3 months using OpenCV and TensorFlow.
• Conducted a review of AI-based retinal imaging telecare services in India to enhance clinical outreach.

Context-Sensitive Semantic Search Tool

• Developed a semantic search tool for impact analysis in software testing, achieving 95% accuracy.
• Generated embeddings using a pre-trained language model for semantic mapping of test cases.
• Built an interactive visualization tool for search results using Python, t-SNE, and matplotlib.