Developed React UI components for GPT-like chatbot application for in-store associates to assist customers. Added backend integrations with Google Gemini Agentic flow via prompt engineering and tool calling.

Integrated Pendo user analytics to provide insights on how users are interacting with specific features of the application.

Deployed computer vision models to the cloud (Vertex AI) for various in‑store applications and services powering the daily analysis of 10's of thousands of images across thousands of stores with low latency

Combined GCP cloud services such as Cloud Storage, Pub/Sub, Vertex AI, GKE, Artifact Registry, BigQuery along with AI models and Python web frameworks to create RESTful APIs and asynchronous background processes for computer vision tasks

Integrated multi-modal LLMs (Gemini, GPT-o4), into one-shot object detection pipelines for product recognition

Optimized CV models for deployment on edge compute (TensorRT+ Nvidia Jetson)

Added instrumentation, logging, and unit testing to computer vision code, increasing app observability and robustness

Leveraged GitHub Actions to automate Docker builds and code testing/deployment workflows

Performed end-to-end load testing of APIs for capacity estimation and optimization

Led the creation of deep learning models for robotic perception.

Developed and deployed machine learning models to enable our robots to pick at 95% quality.

Developed and deployed machine learning models to enable our robots to navigate within several farms robustly.

Developed the dataset creation, training, and evaluation section of our MLOps system, enabling rapid turnaround of ML models.

Wrapped Pytorch computer vision models in ROS nodes, creating an on-robot deployment environment for ML models.

Created web-based tools to make ML accessible to other engineers across different teams for debugging and testing.

Enabling perception of fruit picking robot

Developing computer vision models (object detection, classification, 3D computer vision, semantic segmentation) and cloud-based deep learning infrastructure (MLOps) to power the perception subsystem for an autonomous fruit harvesting robot. Building internal web-based tools for model training and evaluation.

Applied state of the art deep learning and robotics techniques such as object detection and 6D pose estimation to food processing. Developed computer vision pipelines that enabled the dexterous robotic manipulation of poultry products. Developed camera calibration software enabling high-fidelity 3D reconstruction.

http://www.fptd.gatech.edu/