Hi! I am Fatema Tuj Johora Faria, currently working as an AI Engineer II at Astha.IT. I design, develop, and deploy generative AI applications with a strong focus on Large Language Models, Large Multimodal Models, and Retrieval-Augmented Generation (RAG). My work spans building LLM-based agents, multimodal AI agents, and multi-agent architectures that automate complex tasks and enable data-driven insights. In my current role, I leverage cloud infrastructure such as AWS S3, EC2, ECR, and App Runner to build scalable, efficient, and reliable AI systems, while also focusing on user-friendly interfaces that make AI interactions intuitive.
Previously, as Senior Application Developer at Dexian (Bangladesh) Limited, I led the architecture of generative AI systems for production-grade deployment using Azure OpenAI, Azure SQL, Azure Blob Storage, and AlloyDB for high-performance vector search, oversaw proof-of-concept development to translate stakeholder requirements into feasible solutions, and mentored junior developers on coding standards and architectural best practices.
Research interests: Large Language Models, Large Multimodal Models, LLM & Multimodal AI Agents, Human–AI Interaction, NLP for Social Good, NLP for Low-Resource Languages, AI in Healthcare, Vision-Language Models, Trustworthy AI, and Computer Vision.
I am open to collaborative opportunities that align with my research goals. Feel free to reach out at fatema.faria142@gmail.com (personal) or ftj.faria@asthait.com (official).
Fatema Tuj Johora Faria*, Mukaffi Bin Moin*, Mohammad Shafiul Alam*, Ahmed Al Wase, Md. Rabius Sani, Khan Md Hasib (* equal contribution)
11th IEEE International Conference on Sustainable Technology and Engineering (IEEE i-COSTE 2025) 2025
Proposes PotatoGANs, using CycleGAN/Pix2Pix to synthesize realistic diseased-potato images for data augmentation, combined with three Explainable AI methods (Grad-CAM, Grad-CAM++, Score-CAM) across three CNN backbones for interpretable disease classification.
Fatema Tuj Johora Faria*, Mukaffi Bin Moin*, Mohammad Shafiul Alam*, Ahmed Al Wase, Md. Rabius Sani, Khan Md Hasib (* equal contribution)
11th IEEE International Conference on Sustainable Technology and Engineering (IEEE i-COSTE 2025) 2025
Proposes PotatoGANs, using CycleGAN/Pix2Pix to synthesize realistic diseased-potato images for data augmentation, combined with three Explainable AI methods (Grad-CAM, Grad-CAM++, Score-CAM) across three CNN backbones for interpretable disease classification.
Fatema Tuj Johora Faria, Laith H. Baniata, Ahyoung Choi, Sangwoo Kang
Mathematics (MDPI), Vol. 13, Issue 18, Article 3046 2025
Evaluates GPT-4o, Grok 3, Gemini 2.5 Pro, and Claude 3.7 Sonnet on remote-sensing VQA using zero-shot, chain-of-thought, and self-consistency prompting (Self-GeoSense), reaching up to 94.69% accuracy on basic judging tasks with Grok 3.
Fatema Tuj Johora Faria, Laith H. Baniata, Ahyoung Choi, Sangwoo Kang
Mathematics (MDPI), Vol. 13, Issue 18, Article 3046 2025
Evaluates GPT-4o, Grok 3, Gemini 2.5 Pro, and Claude 3.7 Sonnet on remote-sensing VQA using zero-shot, chain-of-thought, and self-consistency prompting (Self-GeoSense), reaching up to 94.69% accuracy on basic judging tasks with Grok 3.
Fatema Tuj Johora Faria, Mukaffi Bin Moin, Busra Kamal Rafa, Swarnajit Saha, Md. Mahfuzur Rahman, Khan Md Hasib, M. F. Mridha
International Journal of Disaster Risk Reduction, Vol. 130, Article 105800 2025
Introduces BanglaCalamityMMD, a 7,903-sample multimodal (text+image) benchmark across seven disaster categories, and DisasterMultiFusionNet, which fuses Swin Transformer and mBERT to reach 85.25% accuracy, a 5.35% gain over the best unimodal baseline.
Fatema Tuj Johora Faria, Mukaffi Bin Moin, Busra Kamal Rafa, Swarnajit Saha, Md. Mahfuzur Rahman, Khan Md Hasib, M. F. Mridha
International Journal of Disaster Risk Reduction, Vol. 130, Article 105800 2025
Introduces BanglaCalamityMMD, a 7,903-sample multimodal (text+image) benchmark across seven disaster categories, and DisasterMultiFusionNet, which fuses Swin Transformer and mBERT to reach 85.25% accuracy, a 5.35% gain over the best unimodal baseline.
Fatema Tuj Johora Faria, Laith H. Baniata, Ahyoung Choi, Sangwoo Kang
Mathematics (MDPI), Vol. 13, Issue 14, Article 2322 2025
Proposes a zero-shot chain-of-thought prompting framework that makes vision-language model reasoning explicit for medical VQA; on PMC-VQA, Gemini 2.5 Pro reaches 72.48% accuracy, ahead of Claude 3.5 Sonnet and GPT-4o Mini.
Fatema Tuj Johora Faria, Laith H. Baniata, Ahyoung Choi, Sangwoo Kang
Mathematics (MDPI), Vol. 13, Issue 14, Article 2322 2025
Proposes a zero-shot chain-of-thought prompting framework that makes vision-language model reasoning explicit for medical VQA; on PMC-VQA, Gemini 2.5 Pro reaches 72.48% accuracy, ahead of Claude 3.5 Sonnet and GPT-4o Mini.
Fatema Tuj Johora Faria, Mukaffi Bin Moin, Zayeed Hasan, Md. Arafat Alam Khandaker, Niful Islam, Khan Md Hasib, M. F. Mridha
International Journal of Information Management Data Insights, Vol. 5, Issue 2, Article 100347 2025
Introduces the MultiBanFakeDetect dataset and MultiFusionFake, an early-fusion text+image model (DenseNet-169 + mBERT) for Bangla fake news detection, reaching 79.69% accuracy versus 73.13% for a text-only baseline.
Fatema Tuj Johora Faria, Mukaffi Bin Moin, Zayeed Hasan, Md. Arafat Alam Khandaker, Niful Islam, Khan Md Hasib, M. F. Mridha
International Journal of Information Management Data Insights, Vol. 5, Issue 2, Article 100347 2025
Introduces the MultiBanFakeDetect dataset and MultiFusionFake, an early-fusion text+image model (DenseNet-169 + mBERT) for Bangla fake news detection, reaching 79.69% accuracy versus 73.13% for a text-only baseline.
Fatema Tuj Johora Faria, Laith H. Baniata, Mohammad H. Baniata, Mohannad A. Khair, Ahmed Ibrahim Bani Ata, Chayut Bunterngchit, Sangwoo Kang
Electronics (MDPI), Vol. 14, Issue 4, Article 799 2025
Presents SentimentFormer, an intermediate-fusion transformer combining SwiftFormer and mBERT for Bangla meme sentiment analysis on the MemoSen dataset, reaching 79.04% accuracy versus 73.31% (text-only) and 64.72% (image-only).
Fatema Tuj Johora Faria, Laith H. Baniata, Mohammad H. Baniata, Mohannad A. Khair, Ahmed Ibrahim Bani Ata, Chayut Bunterngchit, Sangwoo Kang
Electronics (MDPI), Vol. 14, Issue 4, Article 799 2025
Presents SentimentFormer, an intermediate-fusion transformer combining SwiftFormer and mBERT for Bangla meme sentiment analysis on the MemoSen dataset, reaching 79.04% accuracy versus 73.31% (text-only) and 64.72% (image-only).