DANILA
KOZLOV
AI researcher and operator.
Previously, Member of Technical Staff at an AI neolab, leading benchmarking, infrastructure, and multi-agent research.
Earlier: Anthropic, Amazon Web Services, and Cisco.
WORK
Most of the future is undefined. Defining it is part of the work I do.
| // AI RESEARCH | ||
| 2026.03 | Whisker | Closed-loop parametric B-Rep generation with visual grounding. Read more → |
| 2025.10 | HouseView | Floor plans to navigable 3D environments with novel-view rendering. 1st at the Black Forest Labs FLUX Hackathon. |
| 2025.09 | AI-Powered SDLC | Open-source AWS reference patterns: OpenAPI+MCP, KB-MCP, AI-driven incident resolution. GitHub ↗ · Talk ↗ |
| 2025.04 | huh.ai | Real-time multi-lingual speech translation for live calls. Won HackLondon 2025. |
| 2025.03 | Autonomous Skill Discovery | LLM-as-critic for RL skill discovery. Zero/few-shot adaptation via learned skill libraries. UCL DARK Lab. |
| // AI SAFETY & SECURITY | ||
| 2025.11 | Agents We Made Insecure | Critical survey of LLM agent attacks: prompt injection, RAG poisoning, multi-agent trust exploitation. |
| // ROBOTICS | ||
| 2026.05 | Born to Learn | How behaviour shapes body-schema learning. Embodied self-supervised RL on the MIMo infant body. |
| 2024.09 | Film Robotics | Real-time motion control. Demoed at IBC 2024. |
| 2022 | Multi-modal robots | Piano-playing robot at UPenn GRASP Lab. Under Prof. Mark Yim. |
| // GENERAL | ||
| 2024.12 | Acoustic Wave Propagation | 2D/3D wave simulator. Finite-difference, absorbing boundaries, noise modeling. |
| 2024 | Acoustic Levitation | ML pipeline for biological sample sorting. Nature Communications. |
| 2022.11 | COVID Vaccine Model | Distribution modelling for Spain. Adopted into national policy. |
At 12, I helped run a cafe. At 16, I started a small profitable business matching online nannies with families, and founded AYM, a worldwide mathematical modelling competition that brought together 500+ students. My team was twice named #1 in Spain for mathematical modelling; our vaccine distribution proposal was adopted into national policy.
I also ran a few 400-person MUN conferences along the way.
My first proper research was at the GRASP Lab at UPenn (robotics, manipulation, control). From there, a full-ride scholarship to UCL, where I studied Robotics and AI.
Outside the work, I love specialty coffee, art, photography, hiking, pottery, and read a lot of books and poetry. I think a lot about open-ended RL, AI safety, and agents. I care deeply about community and sharing opportunities, this year I organized 3 hackathons, with AgentVerse being one of world's larger agentic AI hackathons.
WRITING
Papers, code, technical notes.
| DATE | TITLE | TYPE |
|---|---|---|
| 2026.05 | Born to Learn: behaviour as a design axis for embodied self-supervised learning | PAPER |
| 2025.12 | Peer stabs you in the back: graceful degradation in mesh networks | NOTE |
| 2025.11 | The agents we made insecure: a survey of LLM agent attacks | SURVEY |
WRITE
ME.
CHANNELS
The future sometimes starts with an email.
- EMAILdanilakozlov.204@gmail.com
- LINKEDIN/in/danilakozlov
- GITHUB@uhDann
- X@iamdanilak