Abhinav Ananthu

OSPP student developer

July '25 - September '25

EmboxRTOS

It's an Open source program funded by many MNCs every year to improve the open source community of different open source projects. The project I worked on aims to integrate Machine Learning Inference Engine like Tensorflow(TFlite Micro) in Embox RTOS. This will allow us to use Embox RTOS for Embedded Machine Learning more effectively. This integration will streamline the development of edge‑AI applications, reduce memory footprint, and leverage Embox’s modular design to support a wide range of microcontrollers and custom hardware platforms.

LFX Mentee

March '25 - August '25

Linux Bug Fixing Program

Linux Kernel Bug fixing operates within a unique collaborative framework that spans the global open-source community.Through hands-on experience with kernel debugging tools, patch development, and community contribution workflows, this program provided invaluable insights into the complexities of low-level systems programming and the disciplined approach required for maintaining critical infrastructure software.

I mainly worked on patches in rust-for-linux project to clean ABI callback functions and porting relevant functions to use sysfs-emit() from scnprintf() function.

IITB - Research intern

May '25 - July '25

Robotics lab , EdgeAi/CV - intern

I developed a custom computer vision dataset of over 15,000 images to detect and classify car fuel system components, managing the full data lifecycle — from collection to annotation — using Roboflow. I performed a detailed comparative study of object detection models including YOLOv8, YOLOv11, and Detection Transformer (DETR), enhancing the YOLOv8 architecture with custom layers to improve small-object recognition. The optimized models were then deployed on OAK-D Pro hardware, achieving real-time inference with integrated depth estimation and spatial coordinate tracking, enabling precise and efficient component localization.

VJTI,COE

January '25 - April '25

ML Research Intern

As a research intern, I worked on multilingual language modeling and transformer-based video understanding. I fine-tuned a large language model (LLM) using Fairseq and MASS on MADLAD-400 and BPCC datasets, improving multilingual performance with a focus on underrepresented languages. In parallel, I designed a Swin-UNet transformer architecture for video key-framing, enabling efficient extraction of key moments from dynamic news feed streams and optimizing temporal representation learning for downstream tasks.