I'm on a mission to build intelligent agents that don't just think—they act. While the world obsesses over bigger language models, I'm engineering the control systems that turn AI thoughts into real-world actions.

🎓 Master's @ Rutgers University | 🔬 Research Assistant @ DIMACS

"The future of AI isn't just about what it can understand—it's about what it can do."

The industry has built incredible AI "brains" (GPT-4, Claude, Gemini), but who's building the body?

That's where I come in. I architect the real-time control stacks that:

• ⚡ Bridge 100ms LLM planning with 60Hz physics control loops
• 🎮 Turn abstract goals into precise motor commands
• 🔄 Learn from interaction, not just supervision
• 🏗️ Scale from simulation to reality

Building hybrid intelligence architectures where:

• GPT-4 decomposes high-level tasks ("make coffee")
• TQC agents execute continuous control (grasp, pour, place)
• Asynchronous pipelines eliminate control latency

Impact: Achieved 50% better generalization on manipulation tasks

Engineering multi-threaded Python systems that:

• Decouple blocking LLM API calls from physics loops
• Maintain stable 60Hz control in MuJoCo environments
• Handle sensor fusion and state estimation in real-time

Designing compositional state representations that:

• Decompose scenes into objects and relationships
• Enable zero-shot generalization to novel configurations
• Published findings on arXiv

Algorithms I Live By: PPO • A2C • TQC • SAC • TD3

while True:
    observe()
    think()  # LLMs for reasoning
    act()    # RL for execution
    learn()  # Continuous improvement

The best AI systems learn from experience, not just data.