Abdur Rafey

Chat with your documents. Ask questions; receive answers alongside the source passage(s) they are grounded in. Know when your document does not answer your question ("context not found" warning). Multiple providers, models, answer languages, retrieval strategies, and document types supported. Configurable temperature and top_p parameters. Enjoy!

A real-time cryptocurrency tracker with a cryptocurrency news feed, also real-time. Production ready. For iPhone & Android.

Aero allows: (1) anyone to rent out anyone's car & (2) anyone to rent their car out to anyone, making it a crowd marketplace for renting cars without the hassle that car rental companies make you go through :')

QMemBN+ is my unified test-time adaptation framework for small vision models that combines quantile-based BN, memory-augmented statistics, and few-shot prototype guidance to adapt on streaming data. It targets real-world distribution shifts with fast, lightweight updates, improving accuracy without retraining or labels.

End-to-end pharma supply chain on Ethereum with role-based batch tracking. Track pharmaceutical products across their supply chain. Know when they're modified, by who, and at what supply stage.

A modular data mining pipeline for multi source Pakistani news, integrating over 600k articles. The system supports automatic section classification using TF IDF baselines and transformer fine tuning, with DistilBERT achieving about 0.84 accuracy and 0.70 macro F1. The workflow is plug and play and easily extendable.

For Lab 1 of Web Engineering at NUST H-12 by Abdur Rafey.

Data structures and algorithms implemented in C++ — searching, string matching, and more, with unit tests.

A real-time cryptocurrency tracker with a cryptocurrency news feed, also real-time. Production ready. For iPhone & Android.

An all-in-one: real-time cryptocurrency market and crypto-relevant news market tracker

QMem-BN is my implementation of a memory-robust, quantile-based BatchNorm layer designed to make small vision models survive harsh distribution shifts without retraining. I combine ideas from Memory BatchNorm (MemBN), Median/Quantile normalization, and Few-Shot Test-Time Adaptation (FS-TTA) into one lightweight, fast, and fully-PyTorch pipeline.

No description.