Profile
Work in progress ...
Education

Computer Science and Engineering

🇳🇴 Erasmus Programme, University of Oslo, Norway

Data Science

🇮🇹 Master Degree, University of Verona, Italy

Banking and Finance

🇮🇹 Bachelor of Arts, University of Brescia, Italy

Experience

AI Researcher

🇯🇵 Current at Equmenopolis - Tokyo, Japan

EQU.ai Developed SynapseLink, a real-time audio-driven facial animation system generating 52 blendshapes at 25 fps (204× faster than real-time, single GPU). Designed a custom training objective for expressiveness (+91.8% over baseline), achieving a 43× lighter model than FaceFormer/CodeTalker. Optimized the production pipeline with tensor caching (57× speedup), mixed-precision training, and streaming inference. Unity & Unreal ready via ARKit.

2025-2026 EU Vulcanus in Japan Fellow

🇪🇺🇯🇵 EU-Japan Centre for Industrial Cooperation - Tokyo, Japan

EU-Japan Centre for Industrial Cooperation For 2025-2026 edition, among 500+ candidates with STEM degrees from all EU, selected as 1 of 23 participants for the industrial internship program "Vulcanus in Japan" (initiative co-funded by the European Commission and the Japanese METI). About the EU-Japan Centre: the Centre is co-funded by the European Commission and the Japanese Ministry of Economy, Trade & Industry. It manages managerial training courses, seminars, researches, student internships and other services all designed to help bring EU and Japanese businesses together. The EU-Japan Centre is member of Enterprise Europe Network.

Data Scientist

🇮🇹 at HPA, Math to Innovate - Trento, Italy

HPA Designed and shipped a suite of AI-native tools for autonomous agents: a privacy-preserving search layer (custom SearXNG backend, now core in several RAG pipelines), an AI-Agent / DB-Agent architecture for low-latency grounded retrieval, and LLM-powered automation pipelines that replaced manual multi-step operations. Owned the full path to production (observability, testing, fallbacks) and led enablement sessions to scale adoption across teams.

Intern as Machine Learning Engineer

🇮🇹 at HPA, Math to Innovate - Trento, Italy

HPA Built BonsaiParser, a modular document parsing framework with smart image filtering and enrichment optimized for RAG pipelines. Fine-tuned WASTE, a computer vision model for waste recognition, applying advanced image processing to improve detection accuracy. Designed a multi-agent system to orchestrate and optimize feature extraction in RAG workflows, improving contextual retrieval quality.

Certifications
  • 2025-2026 EU Vulcanus in Japan Fellowship (one of 23 selected across Europe from 500+ applicants)
  • Check LinkedIn for full list of certifications
Skills
  • NLP
  • LLMs
  • Python
  • PyTorch
  • TensorFlow
  • Deep Learning
  • Dialog Systems
  • Instruction Tuning
  • SQL
  • Leadership
  • Public Speaking
  • Social Impact
  • Italian (Native)
  • English (Fluent)
  • Japanese (Basic)
© 2026 Gabriele Durante. All rights reserved.