Welcome to my GitHub profile! I'm a passionate student pursuing an Integrated Dual Degree in Mathematics and Computing at IIT (BHU), Varanasi. I specialize in AI/ML research, with experience in multi-modal learning, NLP, and computer vision. I'm enthusiastic about developing innovative solutions for real-world challenges.

• Designed a novel two-stage retrieval pipeline leveraging Jina Embeddings-v3 for page-level retrieval and RAPTOR as the second-stage retrieval module.
• Implemented an interleaved reasoning approach for LLMs to dynamically switch between reasoning and retrieval.
• Built a Code and Reasoning Agent capable of handling API failures, debugging code, and retrieving relevant context.
• Achieved robust performance in question answering across diverse domains.
• Tools: Python, LangChain, Pathway
• Blog: Pathway Blog on Multi-Agent RAG
• GitHub Repo: GitHub - Pathway Project

• Developed a custom architecture combining BLIP-2 and Q-Former to effectively capture spatial and temporal features.
• Integrated Llama3.2-1b for advanced reasoning and understanding in video-based question-answering tasks.
• Applied quantization techniques to optimize model performance for large-scale inference.
• Tools: PyTorch.
• GitHub Repo: GitHub - GIF VQA

• Interned at Visual AI Lab, Changwon under the guidance of Prof. Oh-Seol Kwon.
• Gained a deep understanding of different pruning techniques for CNN and Transformer-based models.
• Studied image captioning methods and related evaluation metrics.
• Developed insights into efficient model compression techniques.
• Certificate: Internship Certificate Link

• Developed and enhanced neural network-based approaches for solving Ordinary and Partial Differential Equations (ODEs and PDEs).
• Improved the Lagaris method by using advanced architectures and novel activation functions.
• Conducted experiments on complex systems like Van der Pol and Duffing equations, achieving higher accuracy.
• Tools: PyTorch.
• GitHub Repo: GitHub - Differential Equation Solver

• Languages: C, C++, Python, JavaScript
• Frameworks & Tools: PyTorch, TensorFlow, LangChain, Qiskit, MERN Stack, Git/GitHub

• Understanding the World’s Museums through Vision-Language Reasoning
  Curated a large-scale dataset of 65M images and 200M question-answer pairs for benchmarking vision-language models across visual question answering tasks.
  • ArXiv Link

• Email: [email protected]
• LinkedIn: linkedin.com/in/naitikag
• GitHub: github.com/naitik-2006
• Codeforces: codeforces.com/profile/_naitik