Agentic AI and Multi-Agent Systems
M:3L develops autonomous, tool-using single-agent and multi-agent AI systems for scientific, engineering, and enterprise workflows.
Research laboratory · Yerevan, Armenia
M:3L — Mathematical Modeling & Machine Learning Laboratory
M:3L (Mathematical Modeling & Machine Learning Laboratory) is a research and engineering laboratory in Yerevan, Armenia, working across agentic AI and multi-agent systems, 3D intelligence and computer vision, CAD intelligence and multimodal CAD reconstruction, physics-driven machine learning, differentiable finite-element methods, metamaterials and wave control, and event-sequence and temporal models.
Established in June 2025 at the Institute of Mechanics of the National Academy of Sciences of the Republic of Armenia.
Research
3D Intelligence / Computer Vision
M:3L develops geometric and spatial AI for zero-shot 3D grounding, open-vocabulary detection, scene layouts, and real-world understanding from point clouds or multi-view imagery.
CAD Intelligence and Multimodal CAD Reconstruction
M:3L develops multimodal systems that reconstruct structured, editable, executable engineering representations from point clouds, multi-view images, and text.
Physics-Driven Machine Learning
M:3L develops machine-learning systems constrained and guided by physical laws for computational mechanics and scientific computing.
Differentiable Finite-Element Methods
M:3L uses differentiable finite element methods to connect mechanics simulation with gradient-based optimization and machine learning.
Metamaterials and Wave Control
M:3L studies inverse design of elastic metamaterials, wave-cloaking structures, and programmable wave propagation.
Event-Sequence and Temporal Models
M:3L develops value-grounded event representations that encode what happened, when it happened, and which outcome followed.
Team
Davit Piliposyan
Siemens · Institute of Mechanics · Yerevan State University
David Aznaurov
M:3L · Moscow Institute of Physics and Technology
Danila Rukhovich
Google · Moscow State University
Levon Khachatryan
M:3L
Nerses Alikhanyan
ServiceTitan · Yerevan State University
Robert Gadukyan
Institute for Informatics and Automation Problems of NAS RA
Publications
Z3D: Zero-Shot 3D Visual Grounding from Images
Nikita Drozdov, Andrey Lemeshko, Nikita Gavrilov, Anton Konushin, Danila Rukhovich, Maksim Kolodiazhnyi
A universal pipeline for localizing objects in 3D scenes from multi-view images using zero-shot methods, without geometric supervision or object priors.
Zoo3D: Zero-Shot 3D Object Detection at Scene Level
Andrey Lemeshko, Bulat Gabdullin, Nikita Drozdov, Anton Konushin, Danila Rukhovich, Maksim Kolodiazhnyi
The first training-free 3D object detection framework. It constructs 3D bounding boxes through graph clustering of 2D instance masks and assigns open-vocabulary semantic labels.
TUN3D: Towards Real-World Scene Understanding from Unposed Images
Anton Konushin, Nikita Drozdov, Bulat Gabdullin, Alexey Zakharov, Anna Vorontsova, Danila Rukhovich, Maksim Kolodiazhnyi
The first method for joint layout estimation and 3D detection without camera poses or depth supervision, designed to work with point clouds, posed images, or unposed RGB captures.
cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning
Maksim Kolodiazhnyi, Denis Tarasov, Dmitrii Zhemchuzhnikov, Alexander Nikulin, Ilya Zisman, Anna Vorontsova, Anton Konushin, Vladislav Kurenkov, Danila Rukhovich
A vision-language model that processes point clouds, multi-view images, and text to generate editable CAD programs. It is the first work to apply online reinforcement learning to CAD reconstruction and achieves a near-zero invalidity ratio.
Drivers of Heritage Site Change: A Remote Evaluation Framework
A dynamic, multi-level assessment using computer-vision analysis of historic satellite imagery over monumental heritage sites of Armenia.
News & Events
New Collaboration with YSU
We are partnering with Yerevan State University's Physical-Technical Laboratory and Department of Radiophysics and Telecommunications to build a digital engineering platform for virtual system modeling and diagnostics. The platform will create digital twins of real experimental systems that scientists can query in plain language for fast, accurate answers, powered by machine-learning models trained on thousands of physics simulations. The project is funded by the Higher Education and Science Committee of RA MESCS, with co-financing from the Enterprise Incubator Foundation.
M:3L Lab at METAMAT2026 in Ajaccio, Corsica
Our team joined METAMAT2026 in Ajaccio, Corsica, France. David Aznaurov presented ongoing work with Imperial College London on machine-learning methods for mechanical structure generation and elastic wave control, alongside five days of exchange on acoustic, mechanical, and thermal metamaterials.
M:3L Secures HTI AI Virtual Institute Grant for High-Performance GPU Computing
M:3L has received access to NVIDIA H100 GPUs through the AI Virtual Institute, a program of Armenia's Ministry of High-Tech Industry. This support expands our capacity to design elastic metamaterials and wave-cloaking structures. The research combines differentiable FEM, neural-field optimisation, and generative design to create manufacturable microstructures for broadband Rayleigh-wave cloaking and chiral metamaterials.
BitGn PAC: The Global Autonomous Trustworthy AI Agents Challenge
M:3L Lab is partnering with hackathon.ngo to host BitGn PAC. BitGn PAC is a global challenge focused on building autonomous, trustworthy AI agents. The challenge brings together researchers and engineers to push the boundaries of agentic AI systems with safety, reliability, and trust at the core.
Zoo3D: Zero-Shot 3D Object Detection at Scene Level
M:3L co-authored Zoo3D, accepted at CVPR 2026, an A*-ranked conference. The work introduces the first training-free, open-vocabulary 3D object detector: it clusters 2D instance masks across views into class-agnostic 3D boxes, then labels them through best-view selection and view-consensus visual-language matching. Zoo3D supports point clouds as well as posed or unposed RGB images, while its zero-shot and self-supervised variants deliver leading results on ScanNet200 and ARKitScenes.
TUN3D: Towards Real-World Scene Understanding from Unposed Images
M:3L co-authored TUN3D, accepted at ICRA 2026, an A*-ranked conference. It is the first system to jointly estimate room layouts and detect 3D objects in real scans from multi-view images without ground-truth camera poses or depth supervision. A lightweight sparse-convolutional backbone feeds dedicated detection and parametric wall-layout heads, enabling one model to work with point clouds, posed images, or fully unposed RGB captures and substantially advancing holistic indoor-scene understanding.
cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning
M:3L co-authored cadrille, accepted as an Oral at ICLR 2026, an A*-ranked conference. The model turns point clouds, multi-view images, or text descriptions into editable Python-based CAD programs within a single vision-language architecture. It combines supervised fine-tuning on large-scale procedurally generated designs with online reinforcement learning driven by programmatic feedback, improving geometric accuracy and code validity and achieving leading results across DeepCAD, Fusion360, CC3D, and Omni-CAD benchmarks.
Enhancing Event-Sequence Embeddings for RAG Systems
M:3L is developing a foundational event-embedding model that jointly captures event meaning, temporal progression, and outcome/value signals. The work targets time-sensitive retrieval and agent memory across domains, supporting anomaly and churn detection, next-action and value prediction, causal and counterfactual analysis, and grounded conversational investigation through a Semantic Vector MCP. This work was conducted within the scope of the Faculty Research Funding Program 2025 of the Enterprise Incubator Foundation (EIF).
Contact and collaboration
Yerevan, Armenia[email protected]
M:3L welcomes academic and industry collaboration across its research areas. See the contact and collaboration page.