Portfolio

Github Profile

Linkedin profile

Resume

About Me

• I am Sai Anurag Neelisetty.
• Masters in Electrical and Computer Engineering - specialization in Artificial Intelligence at University of Waterloo, Canada.
• Bachelors in Electrical and Electronics Engineering at Vellore Institute of Technology (VIT), India.
• 3.5 years of professional experience in the software industry [Renault Nissan Technology and Business Centre India (RNTBCI)].
• Experienced in Agile methodologies.
• Passionate about building applications that leverage machine learning to solve problems.
• Skilled in cleaning, analyzing, visualizing, interpreting insights from data and building machine learning models.
• Experienced in automation testing and building applications (Windows standalone, Android).
• Certified Tensorflow Developer.
  

Skills

Languages	Databases	Cloud	ML Libraries	Technologies & Tools	Testing tools
Python, SQL, C #, Java	MySQL, MongoDB	Azure	TensorFlow, NLTK, NumPy, Pandas, Scikit-Learn, Matplotlib	Spark, AzureML, SPSS, Docker, Git, Jenkins, Circle CI, Jira, Flask, Heroku	PyTest, Robot Framework, Appium, JUnit, Cucumber

Experience

Machine Learning Engineer at Renault Nissan Technology and Business Centre India (RNTBCI) from Jan 2020 to Dec 2020.

• Contributed in the development of an automated machine learning (AutoML) web-based application using which employees can request new ML algorithms or use the existing algorithms for their problems.
• Responsible for implementing ML solutions in the backend with Python.
• Developed machine learning prototypes to solve client case studies like predicting product weights to mitigate manual errors, Classifying custom codes of automobile parts to make the manual code assignment faster.

Software Developer at Renault Nissan Technology and Business Centre India (RNTBCI) from July 2017 to Dec 2019.

• Developed Augmented reality/Virtual reality apps for business use-cases.
• Developed new features and maintained the repository of a Windows-based standalone application that interacts with CAN bus and provides a GUI to visualize CAN signals.
• Contributed to the End to End Automation team, where I was responsible for:
  1. Automating the testing of connected car service APIs.
  2. Contributing to the development of Windows and Linux-based test frameworks.
  3. Testing the performance of these services.
  4. Vehicle HMI automation.
     

Projects

• Music2Score: A web application to convert audio files to sheet music. Developed Python scripts to make the conversion in the back end, test scripts, containerized the conversion using Docker and set up continuous integration.
• Statistical Analysis: Performed analysis on a survey data from brand X using SPSS, a statistical software. Applied statistical techniques to analyze, interpret and answer multiple questions on the data to understand the customer purchasing behaviour.
• Image Colorization: Literature review of state-of-the-art algorithms was performed on the image colorization problem. Colorization of grayscale images was then enhanced using transfer learning inspired by EfficientNet-B4 architecture.
• Recurrent Neural Network: Generated artificial poems using word-based and character-based text generation techniques with Bi-directional Long short-term memory Recurrent Neural Network architecture.
• Transfer Learning: Lysine (Protein) methylation sites were accurately predicted for imbalanced data using Convolution Neural Networks inspired by VGG architecture.
  

Research

Image Colorization using Deep Learning: Literature Review

Abstract:

Image colorization is an image-to-image translation problem. It can be considered a pixel-wise regression problem where structures in inputs and outputs are highly aligned. Automatic colorization is an active area of research because of its wide variety of applications. There are several techniques to solve the problem, and the colorization networks used are classified into several categories. This paper aims at studying the state-of-the-art algorithms implemented for the image colorization problem. The work reviews and summarizes three deep learning models with plain network architecture. The inputs to these networks are natural images, and the output is a single image. The primary focus of the papers is on automatic colorization without human intervention.

Image Colorization Enhancement using Transfer Learning

Abstract:

Image colorization is an active area of research, and many new strategies are introduced every day wherein deep learning is extensively used to enhance the process. In this paper, we propose a transfer learning method that uses EfficientNet as a feature extractor to aid the process of fully automatic image colorization. Later, the results of the suggested method have been compared with few existing techniques which aim to solve the same problem.

5G Millimeter Wave – Beamforming Challenges and Techniques

Abstract:

The 5G network is planned to serve a high quantity of mobile data traffic and a significant number of wireless connections while improving communication delay, reliability, and security. Applications such as autonomous cars, remote surgery, augmented/virtual reality games demand an increase in data rates. Millimeter-wave (mmWave) is one of the critical innovations to get a better data rate in 5G communications. It is the under-utilized spectrum and is now used for 5G networks and beyond. But there are challenges with mmWaves as they are prone to high path loss, susceptible to blockages, and fading. Several cellular networks, base stations are required at very short distances to maintain the communication. There are several techniques proposed to mitigate the challenges. As we advance, Massive MIMO combined with beamforming will play a key role in 5G communication and is now a primary research area. Using beamforming, wireless signals can focus on a targeted device instead of spreading from a broadcast antenna in all directions. This will result in a stable, faster, and reliable connection. Fine beam alignment between user and antenna is very crucial for initial access and needs to be maintained owing to the channel fading and user mobility. This paper reviews classification of beamforming that includes analog, digital, and hybrid in detail and their applications in real-time. Also, this paper aims to study various beamforming methods, analyze challenges and identify the best-suited techniques that can be employed in massive MIMO frameworks.

“Maximum power point tracking for PV panels using ant colony optimization” Published in IEEE

Abstract:

Output power of solar panel varies with respect to irradiance and temperature. So maximum power should be extracted from solar panels using Maximum Power Point Tracking. As irradiance changes continuously during the day, many algorithms fail to give maximum power as they get stuck at local maximum. In order to overcome this problem, we propose Ant Colony Optimization which obtains global maximum and improves the efficiency of solar panels. The proposed work has been implemented using Matlab/Simulink. Also, the hardware results obtained indicate the superiority of Ant Colony Optimization over the conventional Maximum Power Point Tracking methods.