Education

• GPA: 3.93/4.00
• Relevant Coursework: Principles of Machine Learning, Multimodal AI, Image and Video Computing, Introduction to Natural Language Processing, Computational Tools for Data Science, Deep Learning
• Relevant Projects: Explainable Stereotype Detection in Text, NASA Airport Throughput Challenge, ISIC - SKin Cancer Detection using Multimodal AI

• Graduated First Division
• Relevant Coursework: Computer Programming, Discrete Mathematics, Machine Learning, Object Oriented Programming, Data Structures and Algorithms, Optimization, Neural Networks and Fuzzy Logic, Digital Image Processing
• Bachelor's Thesis: Unsupervised Event Extraction from Unstructured Data for Carbon Calculation in Recipes

Work Experience

• Revolutionized data processing for LLMs, transforming Excel datasets into a queryable format for seamless integration and enhanced model accessibility.
• Driving a dynamic tech stack transition, orchestrating the migration of our pipeline to state-of-the-art technologies, showcasing proactive leadership in adopting cutting-edge solutions.
• Skills: Python, AWS Services, LangChain, Gradio, FastAPI, Databricks, API Development, Vector Databases

• Supervisors: Prof. Antti Oulasvirta, Mukesh Israni
• Developed and transformed user interface research into a market-ready industrial product, enabling designers to collaborate and co-design with an automated tool.
• Implemented end-to-end model creation as a full-stack developer using technologies like ReactJS, Typescript, Google Protocol Buffers, and Python.
• Conducted in-depth research on machine learning models used as the foundation for the product, gaining valuable insights into their functionality and application.
• Adapted quickly to changing requirements and priorities, displaying flexibility and resourcefulness in a fast-paced, early-stage startup.
• Skills: ReactJS, Django, Git, FigmaAPI, Docker, Google Protocol Buffers, Kubernetes, Python

• Developed a robust content-based recommendation system from the ground up.
• Gained proficiency in utilizing machine learning and natural language processing techniques.
• Skills: Recommendation systems, Flask, Python

• Collaborated on the automation of a drainage system for the Industrial Automation and Control Sector of Rajshree Cement Works, gaining hands-on experience in implementing PLCs (Programmable Logic Controllers).
• Developed a deep understanding of PLC programming languages, such as ladder logic and structured text, to effectively automate the drainage system.
• Skills: PLC Design

• Honed skills in full-stack web development utilizing technologies such as HTML, CSS, JavaScript, AngularJS, NodeJS, and MySQL through successful delivery of the project.
• Gained valuable insights into corporate culture, fostering a strong understanding of professionalism and teamwork.
• Skills: HTML, CSS, JavaScript, AngularJS, NodeJS, MySQL

Projects

• Supervisors: Prof. Surekha Bhanot, Dr. Riza Batista-Navarro
• Utilized Interactive Machine Learning and Machine Reading Comprehension to develop an innovative model for carbon emission calculation in recipes at the University of Manchester.
• Created a system that extracts events from unstructured data using a semi-supervised approach through a combination of human-in-the-loop training and HDBScan clustering, generating machine-readable event templates without pre-existing annotations.
• Enhanced the system’s performance by training models, such as BERT and T5 transformers, to ask relevant questions and refine event details, incorporating POS tagging to improve accuracy.
• Inculcated strong research, problem-solving, and critical thinking skills through collaborative work with academic mentors and experts, enhancing communication, teamwork, and model design abilities.
• Github

• Developed "C-Former," a novel transformer-based architecture for EEG emotion classification, integrating convolutional token embedding layers and feature extractors to enhance spatial and temporal feature representation.
• Achieved state-of-the-art performance on the SEED benchmark dataset, demonstrating improved robustness to data variations and outperforming prior models in accuracy.
• Conducted independent research to pioneer the first application of transformers for EEG-based emotion recognition, contributing to innovation in the intersection of deep learning and neuroscience.
• Github

• Evaluated the performance of SOTA transformers like BERT, RoBERTa, GPT-2, in detecting stereotypes in text.
• Analyzed model behavior on vague and ambiguous sentences to compare performance with human judgment.
• Leveraged attention mechanisms to explain and validate misclassifications using explainability tools like SHAP and LIME.
• Identified significant impacts of dataset underrepresentation and tokenization artifacts on model predictions, highlighting areas for improvement in stereotype detection.

• Designed a multi-modal classifier to classify skin lesions as malignant or benign using image data and tabular metadata.
• Addressed data imbalance by using a combination of data augmentation, SMOTE, and Stratified K-Fold Cross Validation.
• Extracted image features using EfficientNet, fused them with engineered metadata features, and utilized a voting ensemble of LightGBM, XGBoost, and CatBoost to achieve 97% accuracy using AUC metrics.

• Designed a regression-based forecasting system to predict airport arrivals over 3-hour windows using historical throughput data, weather forecasts, and airport configurations.
• Engineered features like rolling averages of weather conditions and delay metrics to capture temporal dependencies and encapsulate key information from noisy, incomplete data.
• Utilized advanced architectures such as Temporal Fusion Transformers and LSTMs to account for time-dependent features.

• Created a model that takes a book title as input, calculates similarity scores between that book and all other books, and returns the top 20 most similar book titles and authors.
• Performed TF-IDF(Term Frequency-Inverse Document Frequency) vectorization using a custom analyzer with word and bi-gram (1-2 word) features, ignoring stopwords, and considering all words.
• Calculated cosine similarity between the TF-IDF vectors, creating a similarity matrix.
• Github

• Leveraged deep learning for automatic music composition generation by analyzing MIDI datasets.
• Extracted musical notes and chords from chord progressions for model training.
• Transformed and preprocessed data to suit a model with LSTM layers, dropout, batch normalization, and dense layers.
• Github

• Conducted research as part of coursework to develop expertise in one-shot instance segmentation, employing Siamese Networks and other advanced concepts.
• Utilized Siamese Networks, FPNs, and ResNet-50 models to implement a system capable of identifying image categories with just a single reference image.

• Conducted coursework involving the simulation of research results in the field of autonomous driving.
• Focused on achieving explainability within autonomous systems, utilizing the BDDX dataset and semantic segmentation techniques.
• Employed Convolutional Neural Networks (CNNs) and visual attention maps to enhance the vehicle's decision-making process

Skills

Certifications

• Introduction to AI and IoT

  IIT Kanpur

• Neural Networks and Deep Learning

  deeplearning.ai

• Scalable Machine Learning with Apache Spark

  Databricks

• Machine Learning in Production

  Databricks

• LangChain-Develop LLM Powered Applications with LangChain

  Udemy

• Data Science, Machine Learning, and Data Anyalysis with Python

  Udemy

Extra-Curriculars

• Actively contributed to the collegiate swim team, participating in both intra-college and inter-college swim events.
• Demonstrated exceptional dedication and skill, resulting in multiple medal wins during competitive swim meets.
• Elected as Vice Captain, showcasing leadership and teamwork abilities.

• Played a pivotal role in the college's Department of Photography, contributing to various aspects of the department's activities.
• Successfully generated revenue for the college fest by capturing and selling portrait photographs of fest attendees.
• Took charge of merchandise design responsibilities for a year, enhancing the departments's branding and overall visual identity.
• Assumed the role of teaching junior members how to effectively use PhotoShop, facilitating their growth and skill development.