Neel Jani

Master of AI candidate with a strong focus on AI for Social Good, combining technical expertise in Computer Vision and Generative AI with a deep understanding of Ethical AI Governance. Proven experience developing rapid prototypes for complex regulatory and environmental challenges in the Aotearoa New Zealand context—from accelerating animal ethics reviews using IBM watsonx to monitoring landslide risks for emergency management. Passionate about leveraging machine learning to solve societal problems while ensuring data sovereignty and privacy.

• Automatic Attendance System, Developed a real-time facial recognition system using OpenCV to identify and log students, demonstrating practical experience in building Identification pipelines.
• Career Point: University Recommendation System, Built a desktop app utilizing a Decision Tree Classifier (95% accuracy) to recommend 1000+ universities based on user profiles.
• Hindi-to-English Speech Translator, Developed a real-time speech-to-speech translation system using ASR and Neural Machine Translation (NMT) architectures.

• Automated Ethics Compliance Tool (Prototype), COMPSCI 714 (Hackathon), Designed to accelerate research funding and compliance for the UoA Animal Ethics Committee (AEC)., Developed a Generative AI scanning tool on the IBM watsonx platform to automate the preliminary review of animal ethics applications., Designed to provide researchers with real-time feedback on application quality, improving funding success rates and ensuring adherence to the '3Rs' (Replacement, Reduction, Refinement) principles., Delivered a functional prototype that demonstrated reduced administrative bottlenecks and faster processing times for university grant applications.
• Geospatial Landslide Detection for Disaster Response, GEOG 761, Developed in the context of Auckland Emergency Management (AEM) requirements., Created a deep learning pipeline to detect landslides in the Auckland region using satellite imagery, aiming to assist Civil Defence in rapid impact assessment following extreme weather events., Fine-tuned a SegFormer (Transformer-based) semantic segmentation model on aerial imagery pre-processed via Google Earth Engine, successfully identifying terrain changes in complex environments., Produced a feasibility study demonstrating how AI can automate situational awareness during climate disasters like Cyclone Gabrielle.
• Privacy-Preserving AI: Machine Unlearning Research, COMPSCI 764, Research on 'Right to be Forgotten' and Data Privacy compliance., Investigated methods to remove specific sensitive data classes from trained deep neural networks without retraining the model from scratch., Explored the SalUn (Saliency Unlearning) framework, implementing Adaptive Thresholding to selectively reset weight parameters responsible for sensitive information., Successfully balanced the trade-off between unlearning efficiency and model accuracy, contributing to the field of Trustworthy AI.

Tennis, Cricket, Volleyball, Music