Lead AI Engineer
At a glance
Senior AI Engineering role at Origin. Bengaluru · full-time.
Pay not stated AI Engineering salary
aimlroles summary, based on the employer's posting.
What you'll do
- Develop diffusion models for simulated wall surfaces, defect examples and adapting models across domains
- Train vision-language and vision-language-action models to connect work orders, CAD plans and sensor inputs with image-level understanding
- Build automated labeling workflows using active learning, self-training and synthetic data across large frame and point-cloud collections
- Prepare models for Jetson-class devices by compressing them and tuning edge deployment under ROS 2
- Carry AI work from problem definition and evaluation through production handoff, while guiding junior engineers and interns
What you bring
- You have at least eight years in deep-learning research and development, or a relevant graduate degree with strong publications
- You have hands-on expertise in diffusion models and multimodal transformers or vision-language models
- You have built large-scale data workflows using active learning, pseudo-labeling or weak supervision
- You are advanced in Python and PyTorch or JAX, with experience in experiment tracking and scalable model training
Who this fits
This senior role suits an AI researcher who can take model work from early investigation into production systems. The position is based on-site in Bengaluru. You’ll also mentor interns and junior engineers.
From the employer
About Origin
Origin (previously 10xConstruction) is building general-purpose autonomous robots for US construction to tackle rising costs, safety risks, and labor shortages. Our modular, multi-trade platform combines purpose-built hardware with real-time site intelligence to navigate complex environments and execute tasks with precision. Trained in high-fidelity simulation and already deployed on live sites, our robots deliver 5x faster execution, 250%+ margin expansion, and significant cost savings. Join India’s most talent-dense robotics team consisting of individuals from IITs, Stanford, UCLA, etc.
About the Role
As a core member of the AI Research team you'll turn cutting-edge, vision-language and diffusion advances into robust real-time systems that see reason and act on dynamic construction sites.
Key Responsibilities
- Research & innovate diffusion-based generative models for photorealistic wall-surface simulation, defect synthesis and domain adaptation.
- Architect and train Vision-Language Models (VLMs) and Vision-Language Action Models (VLA) objectives that connect textual work orders, CAD plans and sensor data to pixel-level understanding.
- Lead development of auto-annotation pipelines (active learning, self-training, synthetic data) that scale to millions of frames and point-clouds with minimal human effort.
- Optimize and compress models (INT8, LoRA, distillation) for deployment on Jetson-class edge devices under ROS 2.
- Own the full lifecycle—problem definition, literature review, prototyping, offline/online evaluation and production hand-off to perception & controls teams.
- Publish internal tech reports and external conference papers; mentor interns and junior engineers.
Qualifications & Skills
- 8+ years in deep-learning R&D or Ph.D./M.S. in CS, EE, Robotics or related field with strong publication record.
- Demonstrated expertise in diffusion models (DDPM, LDM, ControlNet) and multimodal transformers / VLMs (CLIP, BLIP-2, LLaVA, Flamingo).
- Proven success building large-scale data-centric AI workflows—active learning, pseudo-labeling, weak supervision.
- Advanced proficiency in Python, PyTorch (or JAX), experiment tracking and scalable training (PyTorch Lightning, DeepSpeed, Ray).
- Familiarity with edge-AI runtimes (TensorRT, ONNX Runtime), and CUDA / C++ performance tuning.
- Strong mathematical foundation (probability, information theory, optimization) and ability to translate theory into production code.
- Bonus: experience with synthetic data generation in Isaac Sim or robotics perception stacks (ROS2, Nav2, MoveIt 2, Open3D).