Jamal Ansary

I am an AI Researcher at Volkswagen Group of America, Innovation and Engineering Center California (IECC), where I work on multimodal agents and vision-language-action models for automotive systems.

Email  /  Google Scholar  /  LinkedIn  /  GitHub

My research interests lie at the intersection of multimodal foundation models and autonomous agents. I am interested in improving how agents perceive, reason, plan, and interact with complex environments through fine-tuning, distillation, reinforcement learning, post-training, and evaluation. I am particularly interested in grounding and decision making in both digital and physical domains, including computer-use agents, digital assistants, robotics, and autonomous vehicles.

Multi-agent framework for robotic HMI testing using multimodal large language models.

Vision-language models for improved detection and analysis of harmful algal blooms.

Using machine learning techniques to estimate survival rate of COVID-19 patients.

Review of sol-gel process applications.

Swarm robotics is a field of multi-robotics in which robot behavior is inspired by nature.

Short research notes on multimodal agents, grounding, reasoning, planning, and autonomous systems.

Grounding is often treated as finding the right button or object. But in real interfaces, the target may not even be visible yet. To open Settings while inside Navigation, an agent must reason about context, go Home, find the Settings app, and only then act. This note argues that grounding is not just localization, it is reasoning over intent, state, semantics, and action.

What started as a side project evolved into GHOST, Volkswagen's AI-powered platform for automated infotainment testing. As the original proposer and technical lead, I led the development of multimodal agents that autonomously interact with Android automotive systems. The project was highlighted by Volkswagen Group Research & Development as an example of how AI is transforming vehicle development.

What started as a side project during the COVID-19 lockdown evolved into a machine learning study on patient survival analysis and discharge-time prediction. The work was later featured by WTOL News and demonstrated the potential of data-driven methods for clinical outcome prediction.