Profile banner
Alexander Demidko

Alexander Demidko

Engineer, researcher, explorer in AI, 3D, and LLMs

alexander-demidko
London, England, United Kingdom
San Francisco, CA
Joined June 2026
Favicon imageuk.linkedin.com
Favicon imagegithub.com
Favicon imagex.com
Favicon imagemedium.com

Network

1.6K connections
💰
Venture Capital Investors
🌐
Built World AI/BIM
🤝
Talent Acquisition Pros
🤖
AI/ML Founders
📐
...

Summary

Alexander Demidko is a co-founder of Ideatura, a British software startup focused on developing an AI assistant for architects that integrates with 3D CAD software. This venture highlights his entrepreneurial spirit and expertise in leveraging AI for design and spatial computing, bridging the gap between artificial intelligence and architectural visualization. nordic9
He has a strong background in applied research and engineering within the 3D and AI domains, with recent roles at Polycam and Matterport as a Senior Staff Research Engineer. His contributions include discussions on topics like automatic defurnishing of indoor panoramas, and a deep understanding of 3D reconstruction techniques, as evidenced by his summary of "Occupancy Networks" and his acknowledgment in a recent github+2
Alexander possesses significant expertise in machine learning and decentralized systems, stemming from his time as a Machine Learning Engineer at Meta and as Cofounder/Chief Scientist at Cloudless Labs (also known as Fluence Labs). He has authored works on securing decentralized databases, demonstrating a comprehensive understanding of Web3 protocols, distributed systems, and their applications. crunchbase+1
His engineering experience extends to developing ML platforms and leading software teams, as seen in his roles at Stitch Fix as an ML Platform Engineer and Metamarkets as a Lead Software Engineer. His work includes contributions to open-source projects related to machine learning and a demonstrated capability in building complex software solutions. github

Work

Education

Projects

Writing

Geometry Image Diffusion: Fast and Data-Efficient Text-to-3D with Image-Based Surface Representation

September 1, 2024

Acknowledged for valuable feedback and insightful discussions in this research paper, which introduces GIMDiffusion, a novel Text-to-3D model using geometry images for efficient 3D shape representation.

Favicon imagearxiv.org

Securing Decentralized Databases, Part I

May 1, 2019

An article discussing the challenges and approaches to securing decentralized databases in the context of Web3 and the decentralized cloud, exploring topics like IaaS, PaaS, and various computational models (fat client, zero-knowledge proofs, thin client + Merkle proofs, thin client + blockchain).

Favicon imagemedium.com

Occupancy Networks: Learning 3D Reconstruction in Function Space (2019)

January 1, 2019

A summary of a paper detailing an approach for 3D shape reconstruction using occupancy probabilities for continuous 3D points, highlighting limitations of existing methods and presenting a network architecture and experimental results.

Favicon imagehackmd.io