Sadab Shiper

I completed my B.Sc. in Computer Science & Engineering at the Islamic University of Technology (IUT) in October 2025, and then worked as an ML Engineer at Apurba Technologies Ltd. (Oct 2025 – May 2026), building Bengali OCR pipelines, document layout detection systems, and exploring VLM fine-tuning. I am currently a Research Fellow at ARI (Arbitrary Research Institute), working on Advanced Natural Language Processing, Machine Translation, Systems, and AI Agent Evaluation.

I am interested in Computer Vision, Natural Language Processing, and Multimodal AI.

News & Updates

Aug 2026BanglaWild submitted to ACL Rolling Review.
Jul 2026Started as Research Fellow at ARI (Arbitrary Research Institute).
May 2026Completed tenure as ML Engineer at Apurba Technologies Ltd.
Nov 2025Paper accepted at IEEE/CVF WACV 2026.
Oct 2025Started as ML Engineer at Apurba Technologies Ltd.
Oct 2025Graduated with a B.Sc. in CSE from Islamic University of Technology (IUT).
Jan 2025Selected for a 4-month hackathon: Geeky Solutions Learnathon 3.0, sponsored by Brain Station 23.
Dec 20243rd place among ~30 teams at the Datathon, 4th DIU National Data Science Summit 2024.
Dec 20248th place among 108 teams at the Game of Datathon, 3rd KUET CSE National Fest.
May 2024Top 20 among ~83 teams at FOREWARN Disaster Hackathon 2.0.

Publications

Enhancing Vision Language Corruption Robustness using Cross-Distribution & Prompted Denoisers

Sameer Shafayet Latif*, Sadab Shiper*, K. M. Rahiduzzaman Kiran*, Md Farhan Ishmam*, Md Azam Hossain, Abu Raihan Mostofa Kamal, Md Hamjajul Ashmafee

AcceptedIEEE/CVF WACV, 2026 · pp. 5754–5765

PAPER SUPP CODE POSTER

BanglaWild: An In-the-Wild Bengali Scene Text Recognition Benchmark for OCR and Vision-Language Models

Sadab Shiper*, Tawsif Tashwar Dipto*, Mir Md Inzamam, Eshat Tanzeem

Submitted to ACL Rolling Review (August 2026)

Code and data will be available upon acceptance.

ARXIV CODE DATA

Re-Think: Interactive Two-Stage Self-Regulated Reasoning for LLMs

Tawsif Tashwar Dipto, Abdullah, Sadab Shiper, RI Salvi, Amina, H Mahmud

WORK IN PROGRESS

* denotes equal contribution.

Selected Projects

Football Game Analysis — Player & Ball Tracking with Team Detection

Computer vision pipeline detecting and tracking players, referees, and the ball using YOLOv8 + ByteTrack; custom team classification via KMeans jersey-color clustering; visual annotations with ellipses, triangles, and track IDs.

Python, OpenCV, YOLOv8, ByteTrack, scikit-learn, Roboflow, supervision

CODE

SahityaBot — Multilingual Bangla RAG Chatbot

RAG chatbot for Bengali/English queries over HSC'26 Bangla literature using Gemini 1.5 Flash; OCR extraction via pdf2image + pytesseract, MiniLM dense embeddings, FAISS retrieval, CrossEncoder reranking; groundedness/relevance evaluation pipeline with auto-generated HTML reports.

Python, FastAPI, Gemini API, Tesseract OCR, FAISS, sentence-transformers, CrossEncoder, Jinja2, Tailwind

CODE

Bangla News Article Classification

GRU/LSTM/CNN models for Bangla news categorization; GRU: 91.72%, LSTM: 90.99%, CNN: 90.60% accuracy; weighted loss for class imbalance; tokenization and sequence padding pipeline.

Python, TensorFlow, Keras, NumPy, Pandas, scikit-learn

CODE