Cortex AI Lab, University of Dhaka

• Conduct research on agentic AI and collaborative multi-agent systems, focusing on coordinated reasoning, task decomposition, and autonomous decision-making in complex environments.
• Investigate safety and alignment of multi-agent systems: behavior of LLM agents under competitive and survival pressures, adversarial robustness and prompt-injection attack surfaces, and investigation and context-discovery behavior in autonomous web agents.
• Develop vision-language systems for plant-pathology reasoning, building intent-driven visual question answering benchmarks and chain-of-inquiry evaluation pipelines.
• Design and evaluate multimodal VQA pipelines for robustness, interpretability, and trustworthy decision support.
• First-authored papers accepted at NeurIPS 2026, Findings of ACL 2026, and two NeurIPS 2026 workshops, and a dataset paper in Scientific Data.

Data and Design Lab (CARS), University of Dhaka

• Led large-scale analysis of power quality and reliability across multiple power distribution regions in Bangladesh.
• Processed and modeled heterogeneous data sources, including power quality meter data, load and consumption profiles, and industrial and consumer billing records.
• Developed machine learning pipelines for failure prediction, anomaly detection, early fault identification, and load/demand forecasting.
• Built automated data engineering and model deployment workflows for scalable monitoring of power system health.
• Contributed to an electricity-domain conversational assistant that lets users query billing issues, outages, and power quality concerns in natural language.

Robodemy

• Designed and implemented IoT-based robotic systems integrating microcontrollers, wireless communication, and real-time monitoring.
• Conducted structured training on robotics fundamentals, embedded systems, and algorithmic problem solving for competitive and research-oriented robotics.

Tech Topia, Dhaka

• Developed embedded robotic systems using Arduino and ESP32 platforms.
• Designed industrial IoT solutions for monitoring, control, and automation.
• Implemented hardware-software integration for real-time sensing, actuation, and edge-level decision making.

Independent research · ManiSkill / MS-HAB, PyTorch

• Reproduced and evaluated a long-horizon TidyHouse pipeline using 21 pretrained Pick, Place, and Navigate policies, with strict success, reachability, and failure-stage diagnostics.
• Designed controlled handoff experiments showing that robot-state mismatch, rather than object-state mismatch, dominated chained Pick-to-Place degradation.
• Improved full-chain success from 0.7% to 2.2% by tuning global base damping and force limits, validated over three seeds with confidence intervals.

GitHub

• Full Gazebo simulation and ROS 2 control stack covering robot modeling, differential-drive control, sensor integration, and autonomous exploration.

GitHub

• Implemented inverse-kinematics-based leg control, coordinated gait generation, and motion planning for a six-legged robotic platform.

Open-source Python package

• Automated exploratory-data-analysis package with visualization, outlier detection, and correlation analysis.