Mir Nafis Sharear Shopnil

I'm an AI Data Engineer at Technovative Solutions. I work on machine learning systems and agentic tools, with an ongoing interest in research.

Previously, I was a Research Fellow at Fatima Fellowship (September 2024–December 2025) and a Data Science Fellow at Faculty. I hold an MSc in Artificial Intelligence from Queen Mary University of London and a BSc in Computer Science from BRAC University.

Publications & preprints

1. Investigating Knowledge Transfer Across Interactive Dialogue Games

   Filippo Momentè, Mir Nafis Sharear Shopnil, Andrea de Varda, Pavel Merinov, Raffaella Bernardi, Oswald Lanz, Alessandro Suglia, Alessandro Torcinovich

   arXiv:2608.23969 · Preprint · August 2026

   Joint first authorship: Filippo Momentè and Mir Nafis Sharear Shopnil.

   arXiv · PDF

2. Evidence-Grounded Multimodal Misinformation Detection with Attention-Based GNNs

   Sharad Duwal, Mir Nafis Sharear Shopnil, Abhishek Tyagi, Adiba Mahbub Proma

   IEEE/CVF CVPR Workshops (PP-MisDet), 2026, pp. 10994–11003

   Proceedings · PDF

3. MERIT: Modular Framework for Multimodal Misinformation Detection with Web-Grounded Reasoning

   Mir Nafis Sharear Shopnil, Sharad Duwal, Abhishek Tyagi, Adiba Mahbub Proma

   arXiv:2510.17590 · Preprint · 2025, revised April 2026

   arXiv · PDF · Code

4. Reflections from the 2025 EcoHack: AI & LLM Hackathon for Applications in Evidence-based Ecological Research & Practice

   Jennifer D’Souza et al., including Mir Nafis Sharear Shopnil

   EcoEvoRxiv · Preprint · April 2025 · Not peer reviewed

   EcoEvoRxiv · DOI · PDF

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

   Faiza Tafannum, Mir Nafis Sharear Shopnil, Anika Salsabil, Navid Ahmed, Md. Golam Rabiul Alam, Md Tanzim Reza

   IEEE UEMCON, 2021, pp. 358–364

   DOI

Publication details

Selected engineering work

• LLM inference on one L40S

  A practical account of quantization, kernel selection, speculative decoding, and deployment. Includes the benchmark setup and the distinction between serving throughput and real agent performance.

• MERIT implementation

  A modular multimodal misinformation-detection framework combining vision-language models with web-grounded verification.

Writing

• How I Got Qwen3.8-27B from 18.66 to 192.40 tok/s on One L40S
  Updated August 31, 2026 · Hashnode · Engineering
• You're Not a Fraud: Beating Imposter Syndrome in the Age of Vibe Coding
  May 29, 2025 · Medium · Notes
• Understanding Causal Model Induction in Neural Networks for Interpretability
  October 13, 2024 · Medium · Research

All writing

Background

CV and experience

Contact

sharears4077@gmail.com