AlbEval: Albanian Language AI
An independent research project that built an Albanian-language benchmark to test how well leading AI chatbots understand and write Albanian, with fine-tuning to close the gaps for this underserved language.
A software engineer and AI researcher building tools that help people tell what's real online. Creator of VerifAI, an award-winning multimodal deepfake detector.
Giulio Dajani is an Italian software engineer and artificial intelligence researcher based in London. He is studying for an MSc in Artificial Intelligence at King's College London, with coursework in deep learning, computer vision and intelligent systems.
He graduated from the University of Greenwich with a First Class BEng (Hons) in Software Engineering. His final-year project, VerifAI, a multimodal deepfake detection framework, received the university's Best Undergraduate Project award.
Giulio is an AI Automation Ambassador for Tech West England Advocates (TWEA), where he builds AI agents and automations that find funding opportunities and cut routine admin work. At King's he is also a Student Ambassador, leading campus tours and supporting open days. In 2026 he led AlbEval, an independent research project building an Albanian-language benchmark for AI chatbots. From 2025 to 2026 he worked as a Technical Support Engineer at Fluent2 Limited in London. In 2026 he ran an independent digital products business, and he founded the indie game studio behind TheLostKeys. Before coming to London he studied Computer Science at the University of Turin. He has also taken part in academic programmes in China and South Korea.
VerifAI is a multimodal deepfake detector. It looks at a video's frames and listens to its audio separately, then combines the two verdicts into one calibrated score. The whole model runs in the browser, so videos never have to leave the user's device.
Participants used VerifAI to check videos, then rated its usability and how much they trusted it. Most reported moderate to high trust. They preferred seeing a confidence score over a plain real/fake answer.
Training on several datasets reduced overfitting. The model was also checked on 15 real-world videos it had never seen.
Combining video and audio beat either one alone, and caught mismatches like a real video with fake audio. If a video has no audio, VerifAI switches to a visual-only model automatically.
Findings from the survey run alongside the project.
King's College London
Dongguk University, South Korea
University of Greenwich, First Class (84%)
Yancheng Institute of Technology, China
University of Turin, Italy
ITIS Technical College, Turin
Tech West England Advocates (TWEA), remote
King's College London
Fluent2 Limited, London
Amano Group, London
theEword (remote)
Leonardo TEC SNC, Turin
SAET S.R.L., Turin
An independent research project that built an Albanian-language benchmark to test how well leading AI chatbots understand and write Albanian, with fine-tuning to close the gaps for this underserved language.
A puzzle-adventure indie game. As founder, I led a remote development team from concept to production, covering strategy, project management and technical direction.
An independent business selling digital products on Etsy and Gumroad. I handled product development, branding, marketing and commercial strategy.
I worked with students and researchers from several universities to design an AI platform for niche, practical applications.
Everything above, condensed onto one page. Download it for recruiters, applications or collaborations.
↓ Download CV (PDF)Open to AI research collaborations, internships and interesting problems.
dajani.giulio@gmail.com