Aashish Pokharel

Aashish Pokharel

Data Scientist and Machine Learning Engineer

About Me

I am a Data Scientist and Machine Learning Engineer specializing in developing scalable AI solutions. Experienced in designing and deploying end-to-end pipelines, conducting data analysis, and building predictive models with high accuracy. Committed to leveraging innovative technologies to optimize performance and deliver impactful results.

Experience

Data Scientist
Tapestry Inc. (Mar 2025 – Present)

  • Data Scientist in consult for Retail Product Price Sensitivity and Feature Generation team.
  • Led the implementation of the ELT Pipeline to generate data for Machine Learning & Data Analytics for North America & Europe Team.
  • Implemented the CI/CD Pipeline using Jenkins and Apache Airflow for smooth and automated deployment.

Machine Learning Engineer
Fusemachines (Nov 2022 – Present)

  • Led the development of RAG Pipeline incorporated with Active Learning to increase the Accuracy of Information Retrieval from 60% to > 95%.
  • Developed systems to Extract information from financial, legal, and business documents using rule-based systems, Multimodal systems and LLMs with accuracy > 80%.
  • Designed and deployed an Information Extraction Engine for financial documents, integrating multiple models in a scalable pipeline. Improved document processing speed by > 50%.
  • Effectively developed and deployed the cohesive ML pipelines to extract information by combining models such as Document Alignment Model, OCR, Document Structure and Table Structure Recognition Models in a single pipeline.
  • Trained and finetuned unified OCR Model for Multi-language setup (Nepali, English) and Document Structure Recognition Model for financial Documents.

Education

Bachelors in Computer Science and Information Technology
Orchid International College, Tribhuvan University (Nov 2018 – Aug 2022)

  • Grade: Distinction (80.07%)
  • Top of the Class’22
  • Dissertation: Handwritten Digit Recognition from License Plates for Low Resource Language (Nepali)

Research Interests

Efficient Machine Learning: Deep Learning Models (especially LLMs) are extremely powerful but come with high computational, memory, and energy costs. Efficient Deep Learning aims to optimize these models through techniques like compression, pruning, and distillation, enabling them to remain lightweight and scalable without losing performance.

Machine Unlearning: ML models (especially) LLM soak up a lot of information, including information from private and harmful data. Machine Unlearning aims to remove the effect of these data from the weights of the model without having to retrain it.

Technical Skills

  • Language: Python, C, C++, Bash, LaTeX
  • Libraries: TensorFlow, NLTK, Numpy, Pandas, OpenCV
  • ML Frameworks: Scikit-Learn, PyTorch, Streamlit, HuggingFace, Langchain
  • Web Frameworks: FastApi, Flask, Django, RESTFul API, NodeJs
  • Database: MySQL, PostgreSQL, MongoDB, ChromaDB
  • Development Tools: Docker, Git, Github, MLflow, AWS

Honors & Awards

  • AI Fellowship - Fusemachines
  • Top of Class’22 - Orchid International College
  • Full-Ride Scholarship for High School - National Education Board

Licenses and Certificates

  • Micro-degree in Artificial Intelligence Certificates (Issued by: Fusemachines, Jan 2024)
  • DeepLearning.AI TensorFlow Developer Certificate (Issued by: Coursera, Jul 2021)
  • Applied Data Science Specialization with Python Certificate (Issued by: Coursera, Nov 2022)
  • Specialized Models: Time Series and Survival Analysis Certificate (Issued by: Coursera, Jan 2023)