Hukam Singh Rana

AI Engineer with 15+ years of experience architecting cutting-edge machine learning solutions. Led development of diffusion models and GraphRAG systems, significantly enhancing NLP and computer vision capabilities for enterprise clients. Expertise in TensorFlow, PyTorch, and MLOps, with a track record of deploying scalable AI systems on cloud platforms. Uniquely combines academic rigor in AI research with hands-on industry experience, enabling rapid prototyping of novel algorithms. Passionate about advancing multimodal AI and explainable machine learning to drive tangible business outcomes.

• Analysis of mahout big data clustering algorithms, Intelligent Communication, Control and Devices, 2018, Springer, Singapore
• Handwritten bengali numeral recognition using hog based feature extraction algorithm, 5th International Conference on Signal Processing and Integrated Networks (SPIN), 2018, IEEE
• Word prediction using collaborative filtering algorithm, Int J Control Theory Appl, 2016
• Identification of glioma from MR images using convolutional neural network, Proceedings of the Future Technologies Conference, 2018, Springer, Cham
• Partially Visible Lane Detection with Hierarchical Supervision Approach, IETE Journal of Research, 2022, Taylor & Francis

• Stylin - AI Clothing Style Transfer, Leading 5-person team in developing diffusion-based model for real-time style transfer. Early prototype demonstrating improvement in style adaptation quality and retaining fabric patterns. On track to optimize performance 3x, targeting 50% increase in potential market reach.
• Boardboard AI - Multimodal Query Engine, Architecting enterprise AI system to serve accurate results. Implementing GraphRAG to improve retrieval accuracy. Managing 10-person cross-functional team.
• Lane detection and Lane Tracking System, Utilized the concept of self-attention to design the deep convolution network. Used the concept of self-supervised learning to train the model on unlabeled images and transfer the knowledge to the downstream task to segment the lanes in given image.
• Text Summarizer, Train a tokenizer for the domain-based text. Build the text summarizer system using transformer-based architecture based on T5. Fine-tuned it using the LoRA Algorithm. Experimented with abstractive and extractive summarization techniques to optimized the model’s output quality.
• Image Compression Artifacts Removal, Developed a solution to enhance compressed images by removing compression artifacts using guided filters. Leveraged ResNet as the backbone network to segment the main object for precise artifact removal. Achieved good improvement in image quality, demonstrating expertise in image processing and neural networks.
• Image Background Removal, Explored advanced segmentation techniques and employed Mask R-CNN for accurate instance segmentation. Successfully removed image backgrounds with high precision, enhancing image aesthetics and usability. Utilized pytoch to streamline the process, achieving good improvement in background removal accuracy.
• Learning Mathematical Operations using Deep Learning, Engineered a comprehensive dataset for mathematical operations. Formulated mathematical expressions as spatio-temporal sequence forecasting problems. Implemented an LSTM-based model to predict mathematical results, showcasing proficiency in sequence forecasting and model architecture.
• Bone Fracture Abnormality Detection, Processed X-ray images using OpenCV to highlight fracture abnormalities. Implemented a CNN architecture for accurate classification of X-ray images into predefined categories. Achieved 97% accuracy in fracture detection, demonstrating expertise in medical image analysis and classification.