Smayan Kulkarni

My name is Smayan Kulkarni, a Computer Science and Engineering (Data Science) student at Dwarkadas J. Sanghvi College of Engineering. I am deeply passionate about the intersections of Machine Learning, Deep Learning, and Generative AI. With a 9.2 CGPA and a strong foundation in Data Structures and Algorithms, I focus on building scalable, real-world AI solutions.

My portfolio includes several high-impact projects across Computer Vision, NLP, and GenAI:

• Floatchat (SIH 2025): Engineered a cloud-native Azure platform for oceanographic data analysis. By migrating to a Parquet data lake, I achieved a 98.4% cost reduction and 450x faster queries. The system uses a metadata-only RAG pipeline to analyze complicated netCDF files via natural language queries.

• Multimodal DeepFake Detection: Developed a parallel detection system that evaluates images, video, and audio using EfficientNet-B0 and a custom MRI-GAN (U-Net/PatchGAN) to expose high-frequency artifacts. This project, which provides in-depth PDF reports and visual explainability via Grad-CAM, earned 4th Place at CIDECODE 2025.

• Badminton & Cricket Posture Correction: I am conducting research on XAI algorithms and Multi-Stream Hybrid Deep Learning for sports posture correction. I architected a Hybrid TCN + GRU model achieving 93% accuracy, providing users with guided suggestions to improve their form based on specific skeletal inaccuracies.

• Research Paper Analyst & Hybrid Plagiarism Detector: Orchestrated a CrewAI pipeline of 6+ specialized agents powered by Llama 3 and Qwen. It features a hybrid plagiarism engine using dense and sparse search in Pinecone, refined by Cross-Encoders.

• Sentiment Analysis of Multiple Review Types: Built a robust NLP program capable of detecting sentiment on a 1-to-5 scale across various domains like movies and restaurants. The system is specifically engineered to be robust against sarcasm and various talking tones.

• VibeCheck: A Music Recommender System using Collaborative Filtering linked to Spotify and Genius APIs, specifically designed to mitigate the "cold start" problem using a 50,000-song dataset.

Certifications & Expertise:
I hold several industry certifications, including the Machine Learning Specialization (Stanford/DeepLearning.AI), AWS Academy Cloud Foundations, and IBM's Python for Data Science. My technical stack includes PyTorch, TensorFlow, Docker, MLOps tools (DVC, MLFlow), and GenAI frameworks like LangChain and CrewAI.