Sonu Kumar Email & Phone Number

Pune, Mh, In

Vancouver, British Columbia, Canada

-Revolutionized mineral exploration by applying advanced deep learning techniques, improving classification accuracy to 95% through the HORIZON project. Leveraged neural network architectures and reinforcement learning on Databricks for scalable data processing.-Developed the GAIA AI model (Geoscience AI Assistant) for Q&A on geoscience, leveraging GPT-4 and other machine learning models on Microsoft Azure. This resulted in a 40% improvement in data-driven decision-making processes.-Utilized QGIS for spatial data analysis, processing over 500 raster and vector datasets to uncover hidden mineral patterns. This effort contributed to a 30% increase in the discovery of critical metals in covered regions, enhancing exploration efficiency and reducing operational costs by 25%.

Patna, Bihar, India

-Coordinated and managed placement activities, ensuring smooth communication between students and recruiters.-Assisted in organizing placement drives and events to facilitate student recruitment.Collaborated with team members to compile and maintain an up-to-date database of student profiles and placement statistics.-Played an integral part in meticulously organizing the 37th All IITs Placement Committee meeting at IIT Patna.

Patna, Bihar, India

Noida, Uttar Pradesh, India

Noida, Uttar Pradesh, India

Developed a natural language to SQL query conversion system powered by large language models (LLMs), utilizing OpenAI and other open-source LLMs, and using LangChain and agents. Key achievements:-Enabled seamless NL to SQL conversion, improving database query efficiency.-Demonstrated significant expertise in leveraging LLMs for practical applications.

Patna, Bihar, India

Work 1: UAV Person Detection Network DevelopmentDeveloped an Improved Real-time Person Detection Network tailored for UAV imagery, combining CNNs and transformers. Key achievements:-Outperformed baseline RTD-Net and state-of-the-art models in detection accuracy and speed.-Addressed challenges of small object detection and real-time processing.-Enhanced RTD-Net design for UAV scenarios.Work 2: Multi-Label Classification for Legal DocumentationCreated a multi-label classification system for the European Court of Human Rights using deep learning and large language models (LLMs). Key achievements:-Achieved highest F1 Score of 0.796 with LongFormer.-Successfully classified legal documents with high accuracy.-Demonstrated expertise in advanced ML, DL, and LLM techniques for real-world applications.

Master'S Degree, Artificial Intelligence, 8.15

Bachelor Of Technology, 9.18