GHOST: Fast Category-agnostic Hand-Object Interaction Reconstruction from RGB Videos using Gaussian Splatting
🏆 1st-place solution in HANDS@ICCV25 workshop on ARCTIC Bi-CAIR benchmark
Publishing name: Ahmed Tawfik Aboukhadra
I'm a researcher at DFKI and a PhD student nearing completion at RPTU. My research revolves around 3D reconstruction, specifically teaching computers to understand how hands and objects interact in 3D space using Transformers, Gaussian Splatting, GCNs, 3D CNNs, and Foundation models. Beyond the lab, I'm deeply enthusiastic about practical AI. Whenever I encounter a problem, I love utilizing AI tools and coding agents to rapidly build solutions. I hold a Master's in Artificial Intelligence from Maastricht University, where I was awarded the Holland High-Potential Scholarship, and a Bachelor's in Computer Science from the German University in Cairo.
Download CVFocusing on 3D reconstruction of hand-object interactions using advanced deep learning techniques.
Leveraging AI coding agents and modern prompting to rapidly build solutions for everyday problems.
From a CS Bachelor's to an AI Master's on scholarship, now wrapping up a PhD in 3D Computer Vision.



🏆 1st-place solution in HANDS@ICCV25 workshop on ARCTIC Bi-CAIR benchmark
AI-powered apps built to solve real-world problems — from idea to deployment.
AI Cinematic & Educational Video Factory. A CLI pipeline that orchestrates Gemini, Wan2.2, and ElevenLabs to produce audience-aware cinematic videos from any topic — with per-scene checkpointing, multi-language narration, and automatic post-production assembly.
A Telegram-first personal AI assistant that listens to voice notes, transcribes them with Whisper, routes intents with an LLM, extracts tasks and events, and stores memory in a vector database. Send a voice note and get structured tasks, calendar events, and searchable memory — all through Telegram.
Automated job scraper and matcher that pulls listings from 7 boards simultaneously (LinkedIn, Indeed, Adzuna, Glassdoor...), scores them against your skill profile using TF-IDF matching, and presents ranked matches in a clean web dashboard with apply/hide actions.
Fast footballer name recognition pipeline for the popular Arabic "T30 Mazad" game. Transcribes rapidly spoken player names using Whisper with ASR biasing, matches against a Wikidata/FBref knowledge base, then verifies constraint questions ("Did they all play for Barcelona?") with rule-based or LLM checking.


Feel free to reach out for collaborations, research opportunities, or just to say hello!