Unreal Labs
Building AI agents for performance marketing.
Aleksei Naumov
Full-stack engineer and entrepreneur building end-to-end AI-powered products.
London, UK
Work
Terra Quantum — Lead AI Engineer
At Terra Quantum, I led a research team working on efficient LLM inference.
Back in 2023 we published at IEEE MIPR and in a Q1 journal, and open-sourced our compression algorithm and compressed GPT-2 weights.
After that, I led the team that built an AI voice agent, in 2024, before agents went mainstream.
- Papers
- TQCompressor (IEEE MIPR), TetraOpt (MDPI, Q1)
- Weights
- huggingface.co/tq-ag/TQCompressedGPT2
- Code
- github.com/terra-quantum-public/TQCompressedGPT2
Literally Me
AI meme face-swap Telegram bot with 150K+ MAU and 1.5M+ users.
UnrealMe
Consumer iOS app for personalized AI videos from selfies, built in 2024 before GenAI video models matured. Built with custom FLUX LoRAs for personalized image generation; I owned the full cycle from product development to the ML/AI backend pipeline for training LoRAs.
Literally Me Photo
AI photo generation Telegram Mini App built end-to-end.
Selected papers
TQCompressor: improving tensor decomposition methods in neural networks via permutations
IEEE MIPR
GPT-2 compression; 1.5× smaller with no quality loss and 30× less training data.
paper · arxiv · hugging face · github
Tensor Network Methods for Hyperparameter Optimization and Compression of Convolutional Neural Networks
MDPI Applied Sciences, Q1
TetraOpt: CV model optimization; compressed ResNet-18 by 14.5× with minimal accuracy loss.
Projects
Whisker / HouseViewer
1st place at Black Forest Labs & NVIDIA London Hackathon; turns floor plans into immersive 3D homes.