Currently serving as a Teaching Assistant for Machine and Deep Learning (Master’s), Machine Learning for Graph Data (Master’s), and Sequential Decision Making (Master’s).

DAISI (Delft AI Safety Initiative) is a student-run community that aims to empower students and academics at TU Delft to reduce the risks from advanced AI Systems. Since our conception in November 2022, we have:1. successfully run 6 iterations of the AI Safety Fundamentals Alignment Course.2. conducted various Apart AI Safety Hackathons and speaker events.3. organized dozens of upskilling workshops and deep dives in AI Safety related topics.Through our efforts, we have built a community of over 40 students and we continue to support, connect, and educate them for a career in AI Safety. Here is my journey with DAISI:- I completed the AISF Fundamentals course, organized by DAISI, in the fall of 2023. I also actively participated in deep dives, book clubs, and hackathons during this time.- I joined as a Board Member in January 2024.- Since then, I have facilitated two AISF fundamentals myself and have been involved in the group's planning, recruiting, and engagement efforts.- I received funding from Open Philanthropy to run DAISI.

1. Recruited and managed a team of eight to develop a recognition system that accurately identifies registered animals from a database based on their nose prints, leveraging cutting-edge techniques in computer vision and deep metric learning.2. Designed and implemented end-to-end data processing, annotation, and quality assessment pipelines, collecting data for over 2000 unique animals across 10 animal hospitals and clinics in India and Japan.3. Implemented a configurable deep metric learning pipeline supporting ResNet-based feature extractors and triplet loss optimization, with extensive data augmentation and mining strategies; integrated support for real-time tracking of embeddings, model convergence, and performance metrics.4. Successfully built a proof-of-concept for internal beta testing.

1. Built and deployed production-grade end-to-end information extraction pipelines using a combination of OCR tools like (Textract and Google Vision) and state-of-the-art Natural Language Processing (NLP) models.2. Implemented adversarial training and robust optimization techniques that resulted in an improvement of ~3% in the error rate of client models.3. Applied transfer learning to finetune pre-trained transformer-based architectures for user-specific use cases.

About SpotDraft: SpotDraft is an AI-driven, end-to-end contract automation platform that helps organizations create, manage and analyze contracts seamlessly using machine learning. 1. Implemented Transformer-based architectures which improved the previously best-performing models (CNNs) by ~20%. Pre-trained transformers from scratch on legal contracts.2. Implemented transformer attention-head pruning which reduced model sizes by ~15%.3. Experimented with induction networks for few-shot classification, entailment-style zero-shot classification models, multi-task architectures, explainable models, and robust optimizations.4. Spearheaded the building of a nuanced text segmenter for legal contracts to extract parent-child text pairs and achieved an error rate of less than ~2.5%.5. Led an initiative to leverage internally built microservices to extract entities from contracts in order to create a Knowledge Graph containing >200,000 entities.6. Built a Human-in-the-loop system that extracts metadata from legal contracts and uses human interaction to validate the model outputs to create a self-improving, continuous learning training loop.

1. Assisted in the preparation of lecture materials and ran laboratory sessions for the Probability and Statistics course.2. Held weekly office hours for doubt clearing and other assistance.

1. Worked with NYU Stern Professor Vishal Singh to build an online repository of interactive graphs and dashboards for interactive country comparison.2. Worked extensively with R and Excel to acquire and clean publicly available macro and microdata. 3. Designed and developed interactive visualizations using Tableau and R (ggplot, plotly, highcharter, leaflet, scatterD3).