• Integrating multimodal RAG into a domain specific LLM on Cardiology to generate differential diagnosis for radiology reports by performing visual Instruction tuning on LLAVA - Med LMM using LLaMa 3.1• Led the development of a machine vision-based method to estimate head pose. Curated a dataset with RGBD & gyroscopic data for accurate head pose estimation using 6DRepNet. Benchmarked estimated range with error < ±5°.• Orchestrated the implementation of CI/CD pipeline to collect and document previous work using Docker on autonomous wheelchair project. Currently incorporating visual SLAM using Intel RealSense D455 to enhance spatial awareness.

Teaching Assistant for the course INST-126- Introduction to Programming for Information Science. Responsible for Assignment Creation, Office Hours, Lab sessions in addition to the general TA Responsibilities like grading and query solving.

Worked at the Perception Robotics Group over the Summer of 2022 on a novel project to develop a multi-modal AI Violin Coach. Exhaustively analyzed and processed 3D spatial data captured from a 15 Vicon motion capture cameras coupled with 7 RGB and audio streams from Phone and Go-Pro cameras to reconstruct actual and virtual 3D entities & features. Produced 18*4 + 6 motion Features to be input to a multi-modal deep learning neural network that uses audio and motion features synchronized with each other with raw, normalized, and encoded versions using basic 3d geometry and mathematics.

Worked as an Intern at MG motors as part of the MG Nurture 2.0 program. 109 applicants across 79 Indian cities were selected for this.

Developed prototypes using Single Board Computer based Smart home security system to detect entrants of a home and learn about the usual occupants locally on board, keeping the data secure and localized. The system used a normal webcam and display an output stream of detected faces in real time with a frame rate as high as 10fps. Deployed the system on multiple configurations centered around different computing environments like Raspberry Pi 3B+, Nvidia Jetson Nano, Google Coral Accelerator & Dev Board. Added text and email alerts for unknown entrants.

As interns, we were tasked with designing and fabricated PCBs of dev-boards using Espressif Systems' ESP-8266 models. We used Eagle 7 for designing the boards. We were also supposed to solder SMD components onto the boards.