Mir Nafis Sharear Shopnil

Data Scientist & AI Researcher

London, UK

sharears4077@gmail.com

mirnafissharearshopnil

MirNafisSharearShopnil

About Me

about.sh

$ cat summary.txt

Hello! I am an AI Engineer and Researcher with a genuine passion for AI. My work has centered on misinformation detection, probabilistic machine learning, and large language models. I have built solutions that merge academic insight with practical applications. I am seeking roles as a Data Scientist or Machine Learning Engineer. I also look for PhD opportunities that focus on building AI models, probabilistic ML, and foundational models. My journey blends deep research with hands-on industry experience. I value clear results and practical innovation.

Education

education.sh

$ cat education.json

Queen Mary University of London

London, UK

MSc in Artificial Intelligence

Sept 2022 – Sept 2023

Grade: 2:1

BRAC University

Dhaka, Bangladesh

BSc in Computer Science

Jan 2017 – Oct 2021

CGPA: 3.45

Honors: Vice Chancellor's List (Summer 2020, Fall 2020), Dean's List (2017)

Professional Experience

experience.sh

$ cat work_experience.log

Machine Learning Fellow

Fatima Fellowship

Sept 2023 – Present

Remote, US

  • Conducted advanced research on misinformation context generation using text and images, employing proxy agents and Large Vision Language Models (LVLMs) like CLIP and BLIP.
  • Developed and fine-tuned multimodal transformer models with frameworks such as PyTorch and TensorFlow, enhancing the accuracy of context generation in preliminary trials.
  • Collaborated with a multidisciplinary team to integrate LVLMs into misinformation detection systems, contributing to improved analytical capabilities and co-authoring a research paper submitted for publication.

Data Scientist (Contract)

Faculty AI

May 2023 – July 2023

London, UK

  • Optimized supply chain operations by implementing predictive models for supplier reliability and ex-factory dates using 20TB of data in Azure SQL Server.
  • Applied techniques such as XGBoost, LightGBM, Bayesian Ridge, and neural networks, achieving high accuracy with a 2% MAE after hyperparameter tuning.
  • Improved operational efficiency, resulting in a monthly labor savings of 500 hours.

Software Developer

Miguns Technologies Ltd.

July 2021 – Aug 2022

Dhaka, Bangladesh

  • Led development of machine learning solutions for fintech applications: fraud detection, payment analytics, credit scoring.
  • Implemented ML techniques (Random Forest, Gradient Boosting, ARIMA, Prophet, neural networks, NLP models).
  • Enhanced customer experience by automating 40% of inquiries, reducing response time from 15 to 2 minutes.

Research Experience

research.sh

$ cat research_experience.md

Research Fellow

Fatima Fellowship

Sept 2023 – Present

Project: Generating Context for Misinformation using Texts and Images using Large Vision Language Model (LVLM)

  • Developed a framework to generate accurate context for misinformation using Large Vision Language Models (LVLMs) and agentic workflows.
  • Implemented chain-of-thought and tree-of-thought reasoning for context generation.
  • Aiming to improve the ability of LVLM models to detect and contextualize misinformation more effectively, contributing to interpretable and reliable NLP systems in real-world scenarios.

Undergraduate Researcher

BRAC University

Jan 2021 – Nov 2021

Project: Demystifying Black Box Learning Models for Rumor Detection in Social Media

  • Developed a hybrid model combining XGBoost, SVM, Decision Tree Classifier (DTC), Extra Trees Classifier (ETC), and Random Forest Classifier (RFC) for rumor detection on Twitter, achieving a 5% improvement over the base model and an accuracy of 93.22%.
  • Investigated model interpretability using LIME, providing detailed insights into feature contributions for predictions.
  • Published findings at the 12th IEEE UEMCON Conference.

Projects

projects.sh

$ ls -la ./projects/

AI-Powered Ecological Assistant with Agentic Workflow

  • Developed a multi-agent AI system (Planner, Evaluator, Executor) to autonomously route ecological queries across tools like GPT-4o, BioTrove-CLIP for species classification, and Wikipedia/RAG pipelines.
  • Engineered a context-aware workflow using Streamlit (frontend) and FastAPI (backend), integrating dynamic tool selection (image/PDF analysis, GPT synthesis, or Wikipedia verification) based on conversational history.
  • Open-sourced the project with modular agent/tool abstractions, enabling reuse in conservation tech or AI agent frameworks.

Skills

skills.sh

$ cat skills.yaml

Programming Languages

Python R C++ SQL

Frameworks and Libraries

PyTorch TensorFlow Scikit-Learn Hugging Face LangChain Unsloth

Machine Learning

Classification Regression Clustering Deep Neural Networks Statistical Modeling

MLOps

Model Deployment Docker MLflow Flask FastAPI CI/CD Git AWS Azure

Large Language Models

Training and Fine-Tuning LoRA RLHF GRPO Retrieval Augmented Generation (RAG)

Databases

PostgreSQL MySQL

Publications

publications.sh

$ cat publications.bib

Demystifying Black-box Learning Models of Rumor Detection from Social Media Posts

Tafannum, F., Sharear Shopnil, M. N., Salsabil, A., Ahmed, N., Rabiul Alam, M. G., & Reza, M. T.

Proceedings of the 12th IEEE Annual Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON), New York, NY, USA, pp. 0358–0364.

DOI: 10.1109/UEMCON53757.2021.9666567

2021

This research developed a hybrid model combining XGBoost, SVM, Decision Tree Classifier, Extra Trees Classifier, and Random Forest Classifier for rumor detection on Twitter, achieving 93.22% accuracy. The study also investigated model interpretability using LIME to provide insights into feature contributions for predictions.

Contact Me

contact.sh

$ echo "Start a conversation with me"

Ready to connect? Click the command below to start an email: