Muhammad Shoaib Lashari

At SoloGenAI Pvt. Ltd., I collect, clean, and preprocess Urdu health-related text for LLM fine-tuning. I also fine-tune models and assess their output quality on domain-specific data as part of a remote team.

For my research, “Beyond Accuracy: Explainability of Transformer Models,” I fine-tuned UrduBERT to a 79.8% F1-score and built a hybrid attribution pipeline to explain its predictions. The work received the Best Paper Award at CSET'26.

I built a University RAG Assistant with LangChain and FAISS to answer university queries from institutional documents using grounded retrieval. I also fine-tuned Stable Diffusion XL for product photography and built an image curation and captioning pipeline.

At Aror University, my additional research includes temporal deepfake detection, Sindhi text-to-speech, and sign language and facial emotion recognition. I hold Google Data Analytics and Google AI Professional certificates, and I enjoy making technical findings clear and interpretable.