.png&w=256&q=75)
2
2
France
2+ years of experience
Research Master student from IP Paris, with sufficient industrial experience in building machine learning systems, along with a solid theoretical foundation in AI. Aims to contribute to Research & Development in the field of AI by publishing papers while actively engaging in state of art projects and collaborations.

Mind Search is a novel, offline contextual information search tool that lets you find files using direct or indirect clues about their contents—no exact keywords, filenames, or tags required. Everything runs locally on your ARM64 device (with NPU acceleration), so your private data never leaves your machine. Unlike Microsoft Recall—which periodically captures screenshots and depends on visual snapshots—Mind Search embeds the actual content (text, transcribed speech, OCR’d image text) into a unified vector space and leverages those embeddings to match even indirect contextual hints, all without any cloud connectivity. - Contextual Content Retrieval: Locate files by supplying direct excerpts (“the paragraph that mentioned GDPR compliance”) or indirect hints (“that contract I reviewed last month about AI ethics”). 100% Offline & Privacy‐Preserving: All indexing and search operate on-device; no internet, no cloud. - Multi-Modal Indexing: Supports PDFs, slides, code, images, and audio in one seamless index. - NPU-Accelerated Embeddings: ONNX-quantized models optimized for Snapdragon X Elite power real-time searches. - Instant Refresh: Simply rerun the indexer to pick up new or updated files—no slow re-indexing cycles.
8 Jul 2025

TopoGen3D is an innovative pipeline that converts 2D images of shapes (e.g., torus, Klein bottle) into fully-functional Open3D code. By leveraging a fine-tuned CLIP model for shape detection and IBM’s Granite-3B-Code-Instruct-2k for code generation, TopoGen3D automates the process of creating 3D models from 2D inputs. The pipeline identifies the best-matching shape based on accuracy scores, generates Open3D code, and enables visualization in web or desktop environments using PyVista. Designed for topology-aware applications, TopoGen3D is ideal for researchers, educators, and developers working in computational geometry, 3D modeling, and visualization. Its modular design allows seamless integration with tools like CLIP, IBM Granite, and Open3D, making it a versatile solution for transforming 2D concepts into interactive 3D models.
23 Feb 2025