Research Projects
Flex-Corridor: Re-configurable Traffic Corridors for Connected and Automated Vehicle Testing @ Argonne National Laboratory
- Developed and deployed C-V2X communication infrastructure using Cohda MK6 OBUs/RSUs, enabling vehicle-to-infrastructure and vehicle-to-vehicle messaging and validating reliable transmission and reception of BSM, SPaT, and MAP data.
- Designed and simulated connected intersection environments by generating custom MAP/SPaT sequences, creating vehicle drive traces, and visualizing traffic signal behavior for controlled Hardware-in-the-Loop testing.
- Built custom data processing and logging tools, including a Python-based hex decoder, timestamp-synchronized CSV logging, and C-based message repackaging pipeline for real-time vehicle state monitoring and radio validation.
- [PDF]
Cyberbullying Dataset
- Coming soon!
- Constructed a labeled dataset for cyberbullying.
- This work is currently under submission.
Network Science for Cyberbullying
- Coming soon!
- Modeled evolving cyberbullying sessions as dynamic directed graph
- This work is currently under submission.
Visuomotor Behavior Cloning with Human Action Reconstruction
- Designed and executed structured data collection pipelines for both human video demonstrations and vision-based robot teleoperation, enabling robust training across four manipulation tasks (Stack, Sort, Duck, Sweep) using the XArm robotic platform.
- Conducted joint training using visual and proprioceptive input modalities, applying retargeted 3D hand pose estimation and action mapping techniques to enable robust, sample-efficient robot policies
Adversarial Strategies and Countermeasures for Cyberbullying Detection
- Designed and implemented end-to-end machine learning pipelines to evaluate cyberbullying detection models, incorporating text standardization, cross-validation, and baseline benchmarking across traditional ML and deep learning models.
- Conducted black-box adversarial attacks using word-level substitutions and character-level perturbations to assess model vulnerabilities and quantify performance degradation.
- Analyzed limitations of small expert-labeled datasets and restricted metadata access, synthesizing findings to highlight challenges in real-world cyberbullying detection
- [PDF]
Dost: Culturally Responsive Support for Central and South Asian Newcomers Using AI
- Built an interactive web-based platform for Afghan refugees to access support resources aimed to reduce stress and enhance integration, including a fine-tuned GPT model offering real-time support.
- Includes fine-tuned GPT assistant, Google Translator API translation from Dari to English, resources for learning English, mindfulness and religious resources.
- [Website]
HeyFriend: Wellness and Resource-Driven Support for Low-Income Chicagoans
- Built an interactive web-based platform for low-income individuals to access support resources aimed to reduce stress and enhance integration, including a fine-tuned GPT model offering real-time support.
- Built a custom resume-rendering tool using rendercv and HTML interface to allow users to create a customized, downloadable, formatted resumé.
- Implemented CareerOneStop API for real-time labor data on job market and certifications, Google Translate API for multi-language translation, and a custom javascript-based real-time data logger tracking user engagement displayed on a live dashboard.
- [Website]
ActionPoint App: Interactive App for Strengthening Parent-Teen Relationships
- Developed ActionPoint, an iOS app grounded in empirical research to strengthen parent–teen communication, set healthy social media boundaries, identify cyberbullying and risk, and reduce negative outcomes associated with online harm.
- Built a cloud-connected system using AWS, SQL, PHP, and Swift (XCode) to securely store responses, sync parent/teen progress, and generate result comparisons.
- Designed an accessible, shared interface enabling linked parent–teen accounts to complete interactive modules and surveys together.
- [PDF]
