cv
An interactive CV, plus a formatted PDF view.
Basics
| Name | Partha Pratim Nath |
| Label | 3D Vision-Language & Spatial AI · Building in UK-EU Defence Tech |
| nath.partha@outlook.com | |
| Phone | +49 15239522871 |
| Url | https://nath-partha.github.io |
| Summary | 3D-vision engineer turned builder — I take frontier spatial-AI research and ship it to the people who need it.
|
Key highlights
-
🛡️ Now — Building in UK-EU Defence Tech (Stealth)
Independent since March 2026 — applying my 3D-perception and spatial-AI edge to UK-EU defence tech.
-
🚀 First to Productise 3D VLM + Autoregressive Scene-Graph Generation
Shipped generative 3D reconstruction to paying BIM/CAD drafters — frontier vision-language and autoregressive scene-graph models turned into a tool real users rely on.
-
📈 Built the Customer Feedback Loop
Reassigned myself from R&D to MLOps to fix the real delivery bottleneck, then took modules straight to drafters — growing usage through fast iteration.
-
📄 ADL4D — Published 4D Dataset
Co-first author of a 1.1M-frame 4D activity dataset (arXiv); designed its weakly-supervised hand-pose annotation pipeline.
-
🏆 Kurt Fischer Prize
TU Munich award for markerless multi-subject hand-tracking research.
-
🚀 Nvidia Inception — Accepted
Accepted into the Nvidia DGX Inception program; led the experiment design and team preparation.
-
📚 Designed a TUM Course
Co-created 'Project Lab: Human Activity Understanding,' now part of the TUM CIT curriculum.
Work
-
2026.03 - Present Spatial AI & 3D Perception
Stealth · UK-EU Defence Tech
Applying 3D perception and spatial intelligence to problems in UK-EU defence tech.
-
2024.12 - Present Machine Learning & Computer Vision Engineer
Cirqular Pointcloud Analytics GmbH
Took frontier 3D spatial-AI research to production — first to productise 3D VLM and autoregressive scene-graph generation for paying BIM/CAD drafters.
- First to ship SpatialLM reconstruction to paying drafters
- Reverse-engineered SOTA training stacks (Meta, HKU); trained VLMs on 60+ GPUs ($150k GCP)
- Moved R&D→MLOps to unblock delivery; closed the customer feedback loop
-
2023.09 - 2024.09 System Engineer — Machine Learning & Computer Vision
RevTec Systems AG · Casinos Austria International
Real-time on-prem surveillance and analytics for casino tables.
- Built and delivered a 3D tracking, analysis & fraud engine for UK & ZA casino chains
- Massively reduced customer onboarding time and costs — from 2 weeks to <5 days (SAM toolkit)
-
2022.10 - 2023.05 Software Engineering Intern
Infineon Technologies AG
Machine-vision software for 3D human pose on a depth camera.
- 3D detection/tracking + multi-view calibration; sensor library for radar gesture training
-
2021.10 - 2022.09 Co-Founder — Assistive Navigation (Dropout)
SmartAIs (Unternehmer TUM)
Founded an early-stage venture to help blind and low-vision people navigate independently — drove fundraising and validation, then wound it down on honest signal.
- Built alongside my TUM motion-capture research
-
2021.06 - 2023.08 Research & Teaching Assistant
TUM Chair of Media Technology
Markerless motion-capture research (Kurt Fischer Prize) and 3D-vision teaching.
- Markerless motion capture from RGB → the dataset that became ADL4D (Kurt Fischer Prize)
- Co-designed a TUM course (now in the CIT curriculum); built a multi-camera capture studio
-
2019.06 - 2020.03 Founder — Multispectral Operator Fusion
Firefighting Drone Venture (SRM)
Led a student venture bringing firefighting drones to South-East Asia — tackling operator cognitive load by fusing thermal, low-light, and RGB into one operator view.
- Secured funding, equipment, and team (4 peers + 2 professors)
- Built 3 MVPs; ran live tests and adversarial trials
Publications
-
2024.02.01 ADL4D: Towards A Contextually Rich Dataset for 4D Activities of Daily Living
Zakour*, Nath* (* equal contribution)
Co-first author. Designed a weakly-supervised markerless hand-pose annotation method for 4D activity understanding under complex multi-subject interaction, and published a 1.1M-frame multi-view, multi-subject dataset.
Education
Skills
| 3D Vision & VLMs | |
| PointTransformerV3 | |
| Sonata | |
| LLaMA 3.2 | |
| Qwen2 | |
| CLIP | |
| SpatialLM |
| Deep Learning | |
| PyTorch | |
| DeepSpeed / ZeRO | |
| HuggingFace | |
| PyTorch3D | |
| Detectron2 | |
| Pointcept |
| Production ML & Cloud | |
| Multi-node training (60+ GPU) | |
| GCP / AWS (A100, L40) | |
| Docker | |
| ClearML | |
| Model deployment |
| 3D Understanding | |
| Point-cloud segmentation | |
| Scan2CAD / Scan2BIM | |
| ICP | |
| SLAM | |
| Multi-view geometry | |
| Camera calibration |
| Programming & Tools | |
| Python | |
| C++ | |
| OpenCV | |
| Open3D | |
| NumPy | |
| Rerun | |
| CVAT |
| Open-Source Datasets | |
| ADL4D (1.1M, co-author) | |
| ScanNet++ | |
| Structured3D | |
| H2O3D | |
| DexYCB |
Languages
| English | |
| Native |
| German | |
| Conversational |
| Other (French, Hindi, Bengali) | |
| Beginner / Conversational |
Interests
| Robotics | |||
| Perception stacks | |||
| Safe grasping & interaction | |||
| Motion Capture | |||
| Marker & markerless | |||
| Monocular & multi-view | |||
| 3D Scanning | ||||
| Point/mesh reconstruction | ||||
| Gaussian splatting | ||||
| Scan2CAD | ||||
| Drone / Autonomous Tracking | |||
| Object & area tracking | |||
| Ground stabilisation | |||
References
| Dr Rahul Gopal Chaudhari | |
| TUM Senior Scientist |
| M.Sc. Marsil Zakour | |
| TUM Doctoral Candidate |
| Maximilian Strobel | |
| Infineon, System Architect — Machine Learning |
| Michael Winking | |
| Infineon, Staff Engineer |