AI & Data Science Enthusiast

Ajhesh Basnet

Hi! I am Ajhesh Basnet, a passionate Artificial Intelligence and Data Science professional dedicated to developing innovative solutions. My expertise spans Deep Neural Networks, Large Dataset processing, with specialized focus on Computer Vision, Natural Language Processing, and MLOps implementations.

Areas of Focus

Machine Learning

Classical ML models, Deep Neural Networks and Model Optimization

Computer Vision

Image Processing and Visual Recognition Systems

Natural Language Processing

Text Analysis and Language Understanding

Data Science

Extracting Meaningful Insights from Structured and Unstructured Data

MLOps

Model deployment, monitoring, and CI/CD pipelines for machine learning systems

Backend Development

Building scalable APIs and services to support AI/ML applications

Notable Projects

Vision Transformer (ViT) – From Scratch Implementation of "An Image is Worth 16x16 Words"

Implemented core ViT architecture including patch embedding, positional encoding, and multi-head attention using PyTorch. Modular design tested on image classification tasks.

Re-implemented Transformer Model (Attention Is All You Need) for English to Nepali Translation

Re-implemented the Transformer architecture in PyTorch for English–Nepali translation, achieving a BLEU score of ~29— comparable to the original paper’s 28.4 on English–French, using fewer configurations.

Low Level Language's Sentiment Classification

Fine-tuned XLM-RoBERTa-3B for 5 class sentiment classification on Nepali text using LoRA, reaching roughly 79% accuracy with balanced training.

RAG-Based Nepali Chatbot enriched with tools calling

Built retrieval-augmented generation chatbot using LLaMA-7B, vector search for domain-specific queries with contextual responses.

SMS Spam-Ham Detection

Developed binary classifier using TF-IDF and Random Forest achieving 93% accuracy with full preprocessing pipeline.

Medical Assistant from Prescriptions

Extracted structured health information using OCR and NER, providing rule-based medical guidance.

Exercise Repetition Counter

Used Mediapipe to track human pose landmarks and count repetitions based on joint movement logic with OpenCV for real-time visual overlay and user guidance.

Real-Time License Plate Recognition and Alert System

Developed a real-time license plate recognition system using a YOLO-based pipeline and OCR, connected to a backend database to detect flagged vehicles and trigger alerts instantly upon match.

Resume

My professional journey combines rigorous academic training with hands-on project experience. The resume provides detailed insights into my qualifications, technical competencies, and notable achievements in the field of AI and data science.

Contact

Connect with me through these professional channels for collaborations or discussions: