Multimodal AI Scientist

Multimodal AI Scientist – L&T Group AI Center of Excellence

Santa Clara, California |L&T Group-wide (US and India)| 8–15 years |

Reports to:

Head, AI CoE

Build multimodal AI systems that reason jointly over vision, language, audio, sensor, time-series, and 3D data to solve real problems across L&T's engineering, manufacturing, infrastructure, energy, and defence businesses. You will bridge foundation-model research and production engineering, working closely with Robotics, Computer Vision, and Language Model teams to build reusable models and pipelines.

• Design, train, and adapt vision-language and sensor-fusion models for engineering drawings, inspection imagery, P&IDs, video, and IoT telemetry.
• Partner with business units on high-value use cases, such as visual inspection with natural-language reporting, drawing understanding, and digital twin enrichment.
• Take proofs of concept through pilot to production, with defined success metrics and quantified business impact.
• Define data collection and annotation strategies, and build scalable pipelines for video, point clouds, and sensor streams.
• Set benchmarks for accuracy, robustness, hallucination, and safety, and optimize models for latency and edge deployment.
• Mentor engineers, contribute to the CoE repository, and publish papers or file patents.

• Define the multimodal AI technical roadmap and reference architecture for the CoE.
• Deliver 3–5 multimodal proofs of concept, at least two scaled to pilot or production.
• Publish reusable model and pipeline assets, plus an evaluation and benchmarking framework.

• Master's or PhD in Computer Science, AI, Machine Learning, Electrical Engineering, or a related field (PhD preferred).
• 8–15 years in machine learning or AI, including at least 4 years in multimodal, vision-language, or foundation-model work and a record of taking models to production.
• Deep expertise in transformers, diffusion models, contrastive learning, and cross-attention fusion.
• Strong Python with PyTorch, JAX, or Hugging Face; distributed GPU training experience on Azure, AWS, or GCP; familiarity with RAG and multimodal retrieval.
• Experience with industrial, engineering, or physical-world data is strongly preferred. Publications at NeurIPS, ICML, ICLR, CVPR, or ACL are a plus.