A comprehensive evaluation benchmark for the Balochi language, covering a suite of natural language understanding tasks analogous to GLUE/SuperGLUE. BGLUE aims to provide standardized metrics for assessing the capabilities of NLP models on this critically under-resourced language, enabling meaningful progress tracking and cross-model comparison for Balochi NLP.

A high-quality instruction-following dataset designed specifically for the Balochi language, enabling supervised fine-tuning of large language models for instruction-following tasks. BIT addresses the critical gap in instruction-tuning resources for low-resource languages, facilitating the development of Balochi-capable chat and task-completion models.

A project focused on adapting Google's Gemma 4 model for Pakistani languages — including Balochi, Urdu, Sindhi, and Pashto — through domain-specific fine-tuning and optimization for edge deployment on resource-constrained devices. The project extends the model with multimodal capabilities to support text, audio, and visual inputs relevant to Pakistani linguistic and cultural contexts.

An agentic AI system designed to automate the process of filing tax returns for Pakistani taxpayers through the Federal Board of Revenue (FBR) portal. TFAgent leverages large language models and tool-use capabilities to guide users through document collection, form completion, and submission, reducing the complexity and time associated with tax compliance in Pakistan.

A multilingual, multimodal mobile navigation agent that provides turn-by-turn directions, local information, and contextual assistance in multiple Pakistani languages. NavAgent integrates vision, language, and location-based reasoning to deliver a culturally and linguistically inclusive navigation experience for users across Pakistan, with a focus on underserved linguistic communities.