A Computer Engineer exploring the future of Artificial Intelligence. I enjoy building AI systems, conducting research, and developing applications using machine learning, deep learning, computer vision, and large language models.
$ whoami
Yamraj Khadka | Machine Learning Enthusiast
$ location
Kathmandu, Nepal 🇳🇵
$ tech_stack
["Mistral-7B", "RAG_Pipelines", "U-Net", "PyTorch"]
$ status
Ready for research collaborations...
$ _
U-NET MEAN IOU
LLM MODEL SIZE
STUDENTS MENTORED
PUBLIC ML DAYS
Planto AI • Remote
Sayapatri Group Pvt. Ltd.
Published in Journal of Engineering Issues and Solutions (Vol 5). This research solves the challenge of urban expansion monitoring in Nepal. I designed a modified U-Net from scratch.
The first Nepal-focused legal LLM fine-tuned on the National Penal Code 2017. This model was adopted and re-quantized by the community into GGUF formats (Q2–Q8) within 24 hours of release, enabling CPU-only inference for low-resource environments.
Mistral 7B LoRA
6,400 Examples
3 Live Spaces
/active_projects/production_grade_ai
Fine-tuned Mistral-7B specifically for Nepal's National Penal Code. Implemented advanced RAG architectures for high-fidelity legal Q&A.
Multi-agent orchestration system built with LangGraph. Includes modules for recursive memory, autonomous reasoning, and safety protocols.
Deep Learning Specialization Insights & Technical Logs
Explored YOLO real-time detection vs. pixel-level U-Net segmentation. Deep dive into Transpose Convolutions for upsampling...
Fine-tuning VGG16 on CIFAR-10. Achieved 73.13% accuracy by freezing base layers and optimizing dense heads...
Applying Andrew Ng's strategy on diagnosing system errors, handling mismatched test sets, and multi-task learning.