Profile
Combining a solid foundation in Mechanical (BE), a deep understanding of Physics (MSc) & expertise in Data Science (BS), passionate about leveraging cutting-edge technologies to drive innovation, with a keen interest in the application of Computer Vision. Expertise in developing AI models and applications using Python, Java & C++. Proficient in machine learning & deep learning algorithms, data preprocessing, feature engineering, and data visualization. Experienced in deploying and scaling AI solutions on cloud platforms such as AWS, Azure & GCP.
Education
BE (Hons) Manufacturing
BITS Pilani | CGPA: 7.64BS Data Science & Application
IIT Madras | CGPA: 7.4MSc Physics
BITS Pilani | CGPA: 7.64Academic Achievements & Awards
Golden Unicorn Physics Association Winner - BITS Pilani (2024)
2nd Prize - Hult Inter-College, BITS Pilani (2022)
1st Prize - Hindi Poem Recitation & English Debate (2013) Bhartiyam International School
1st Prize - Science Quiz Competition (2016) Bhartiyam International School
1st Prize - Great Speech & English Debate (2018) Bhartiyam International School
Research Publication
1- Development of a Hybrid Model to Estimate Surface Roughness of 3D Printed Parts (published in Journal of Additive Manufacturing July of 2024)
2- Development of a Semi-Empirical prediction model to estimate surface roughness of 3D printed parts (under Review)- The research introduces a novel robust semi-empirical prediction model that combines the theoretical knowledge of the additive manufacturing and the measurement process with the empirical data.
Languages
Python: Proficient in developing and deploying data analysis, machine learning models, and automation scripts. Experienced with libraries such as Pandas, NumPy, TensorFlow, and Scikit-Learn.
C++: Skilled in writing efficient and optimized code for performance-critical applications. Familiar with object-oriented programming, memory management, and implementing algorithms and data structures.
Java: Experienced in building robust and scalable applications. Proficient in using frameworks such as Spring and Hibernate for backend development.
HTML, CSS, JavaScript: Adept at creating responsive and interactive web applications. Knowledgeable in modern front-end frameworks and libraries like React and Angular, and experienced in using tools like Bootstrap and SASS for enhanced styling.
Key Skills
· Prog Languages: Python, Java, C++
· AI Frameworks and Libraries: TensorFlow, PyTorch, OpenCV, Scikit-learn
· Machine Learning Algorithms: Supervised and Unsupervised Learning, Deep Learning, Neural Networks
· Data Processing: Data Preprocessing, Feature Engineering, Data Visualization
· Cloud Platforms: AWS, Azure, GCP
· Software Development: Version Control (Git), API Development, Agile Methodologies
Funded Projects
1) 50k funding from the AUGSD BITS UG project program for developing an autonomous material handling robot. I reconfigured all four motors and both the O drives with minimal documentation. Each motor had 72 independent variables, which required determination through various ground experiments. Additionally, I rewrote the Python code to abstract this low-level complexity while still allowing fine control over the motors. I also developed a basic autonomous material handling system using computer vision, incorporating lane following object detection and placement. Furthermore, I facilitated the acquisition of eight Raspberry Pi 5MP 62 FPS cameras and implemented an improved power system for the rover.
Developed an automated data collection pipeline for 3D printed parts using a Mitutoyo Quick Scope. Utilized a Python script to automate data collection and modeled a custom clamp in AutoCAD to automate the X and Y axes (Z-axis was already automated). Conducted stress and strain analysis on the clamp in AutoCAD, iterating three times to produce a functional part. Integrated a servo motor and Arduino board for precise platform movement, and added a feedback loop from a vision module to correct errors from the motor or encoders. This setup enabled the collection of 10,236 images (12GB of data) in 4 hours, forming the backbone of several upcoming projects.
3D Reconstruction Model for Surface Roughness Measurement
Created a 3D reconstruction model to measure the roughness of 3D printed objects, achieving an accuracy of ±0.1 microns, meeting industry standards.
Automated Material Handling Robot
Developed an automated material handling robot that won 5th place in an Indian corporate competition. Configured the motor controller (O drive) and rewrote the Python program for finer motor control. Assisted in integrating eight Raspberry Pi cameras with Jetson Nano to handle simultaneous inputs and inference for object detection.
Tool Wear Prediction Model for Turning and Milling Processes
Assisted in acquiring sensors (vibration, temperature, and acoustic) to develop a tool wear prediction model for turning and milling processes. Developed a multimedia and multi-head deep learning model for prediction, currently ongoing.
2) 42k + Professor's personal funds - developing an online tool for wear measurement and prediction model using multimodal deep learning architecture. also helped in making DOE, SOP of measurement and accusation of multiple sensors (vibration, temperature, acoustics). In this project, we have explored various transfer learning approaches
Other Academic Projects
1) Auto sample (Biggest project) made quality analysis with the ability to check 4.55 kg/sec grams with 10.3 FPS or with 0.1 second inference speed. This machine had the ability to process 393,120 kg grams per day. The whole experiment was conducted by my company (AgNext) had provide us the conveyer belt and we have to set up the camera the after, it we build the data set for model training we have captured more than 10k image then with help of data annotations team we annotated them I personally another 400 images to speed up the process. Applied Deep SORT, a state-of-the-art tracking algorithm, to enhance the performance and accuracy of the chili segmentation model. So, tracking can be considered in the following two-step process
1. We do the detection and localization of the object, using any type of object detector such as YOLOv7, YOLOR. We have used YOLOv3 because for the given computational resources, it was working fine.
2. Second Step, using a motion predictor, we predict the future motion of the object using its past information. After this, we have to retrain the given model multiple times. Finally, this code is used in production code after optimization.
3) Fruit Quality Analysis
Reduced inference time from 1.4 seconds to 0.003 seconds, resulting in a performance improvement of 43,750%.
Achieved this by exploiting controlled environment conditions and using photogrammetry instead of a deep learning model.
Further optimized by implementing the logic in C++.
4) Red Chili Quality Model Improvement
Improved model accuracy from 91% to 98% by implementing a preprocessing layer.
Addressed issues with overlapping chilies by segmenting them, significantly enhancing accuracy.
Conducted hyperparameter tuning for deployment on edge devices like Raspberry Pi.
Developed a chili-specific segmentation model using OpenCV image processing techniques instead of conventional algorithms.
The project involved creating a custom image segmentation algorithm tailored for chili-specific characteristics to run efficiently on constrained hardware.
