Spearheaded the development of a novel stochastic optimization algorithm, showcasing innovation and problem-solving skills.
Got deep understanding of the intricacies of optimization techniques, loss landscapes, convex and non-convex optimization.
Object Detection
Python
07.2023 - 09.2023
Designed and implemented CNN architecture from scratch in Python without TensorFlow, including Conv2D, Pooling, FCLs, Batch Normalization, and Dropout regularization, using the CIFAR-10 dataset
Also implemented the same using TensorFlow.
Accuracy: >88%
Digit Recognition
Python
05.2023 - 07.2023
Without being dependent on any third-party library - Complete design and implementation of a neural network architecture, activation functions, optimization algorithm selection, training on the dataset, learning parameters, reducing cost, and iterating to get the best model.
Achieved over 99.98% Train Accuracy and 98.57 Test Accuracy
8051 UC Assembler
C
09.2022 - 01.2023
Created a robust assembler for 8051 assembly language, converting code into Intel Hex format. Processed (.asm) files, extracting opcodes, labels, assembler directives and operands. Implemented intelligent op-code decoding as per 8051 documentation, producing efficient Intel Hex (.hex) output.
Demonstrates strong attention to detail and expertise in low-level programming.
Quadcopter
Embedded C
07.2021 - 06.2022
Coding of Electronic Speed Controller, Flight Controller, and BMS of a Quadcopter. IMU Data processing, Interrupt handling, sensor interfacing, coding trapezoidal commutation, etc.
Leading the team to the goals set and executing the project on time, within budget, and with quality outcomes.
Optical Character Recognition
Java
03.2020 - 12.2020
No external OCR library used - Fair amount of success in developing a library for OCR to convert image paragraph to text paragraph by using self-developed algorithms for detection– Letter extractor and Path Finder.
The sub-parts included Line detection & separation, Letter separation, character extraction, feature finding (edges/curves) and ultimately running the self-developed path finder algorithm on the features to get the character.
Education
B. Tech - E&TC -
COEP - College of Engineering, Pune
India
06.2025
Engineering Diploma -
Govt. Polytechnic
Pune, India
07.2022
Class 10 - Class 10
Euro School
Pune, India
02.2019
Skills
Neural Networks
Deep Learning
Stochastic Optimization
Optical Character Recognition
Root Cause Analysis
Embedded Software
Publications
Currently submitted and undergoing review and in the processing stage.
ASRO: Sign-Based Optimization for Stochastic Learning, IEEE Transactions on Artificial Intelligence
Asro, the Adaptive Sign Reversal Optimization algorithm, addresses a critical challenge in neural networks’ training time required to attain a specific cost. While neural networks offer rapid inference, their extensive training periods hinder efficiency.
Asro introduces a step size adaptability and reduction mechanism based on the sign reversal of successive first moments aiming to detect local minimum, leading to accelerated training and cost reduction compared to mainstream optimization algorithms.
Certification
Artificial Intelligence: Knowledge representation & reasoning, IIT Madras - Topper
Understanding Design, IIT Bombay - Topper
Neural Networks & Deep Learning, Coursera
Improving Deep Neural Networks, Hyper Parameter Tuning, Regularization and Optimization, Coursera
Key Account & Marketing Manager at Kuvos Tech (A DYNAMIK LABS GROUP OF COMPANY)Key Account & Marketing Manager at Kuvos Tech (A DYNAMIK LABS GROUP OF COMPANY)