Ratna Prakarsha Kandukuri

---

Software Development and Machine Learning Engineer

• LinkedIn
• GitHub
• Email
• Google Scholar
• Medium

• View Resume

About Me

Hey everyone, Prakarsha here 👋🏼

Welcome to my professional portfolio! I am Ratna Prakarsha Kandukuri, a dedicated Machine Learning Research Engineer with a passion for developing innovative solutions. With a strong educational background and hands-on experience in both academic and industry settings, I am committed to leveraging my skills to drive impactful projects.Feel free to explore my portfolio and connect. I'm always open to discussions about potential collaborations and opportunities. Whether you're a fellow enthusiast or a hiring manager seeking talent, I'm excited to connect and discuss how we can make a positive impact together.

Current Research

ML Research Engineer @ VOCADIAN

Utilizing voice biomarkers and circadian science to detect and manage fatigue risk in workforce

• More

Previous Research

T-REMUS MISSION SIMULATOR TO COLLECT DATA AT RIVER PLUMES FROM REGIONAL OCEAN MODELING SYSTEM (ROMS)

• More

Featured Projects

NATURAL LANGUAGE PROCESSING- LLM

Automated Contract Clause Understanding and Risk Assessment Chatbot with fine-tuned Legal-BERT and GPT-4o

This project aims to automate the understanding and risk assessment of contract clauses using a fine-tuned Legal-BERT model for classification and GPT-4 for generating detailed risk analysis and integrated explanations.

• More

Machine Learning

Dog Breed Classification

A machine learning project utilizing Convolutional Neural Networks (CNNs) for accurate dog breed classification from images, achieved through transfer learning and pre-trained models like VGG16, VGG16BN, and Haar feature-based cascade classifiers.

• More

Machine learning deployment- Sagemaker

Sentiment Analysis Web App

A simple web app which interacts with a deployed recurrent neural network performing sentiment analysis on movie reviews. SageMaker is used to construct a complete project from end to end.

• More

SQL

Animal Welfare Management

The animal welfare database model streamlines medical treatments, shelter operations, and pharmaceutical transactions for veterinary clinics, pharmacies, and shelters. It organizes animal data, manages inventory, schedules appointments, and tracks pharmaceutical orders, enhancing operational efficiency.

• More

Machine Learning Deployment

Plagiarism Detector

In this project, I build a plagiarism detector that examines a text file and performs binary classification, labeling that file as either plagiarized or not, based on how similar it is to a provided source text.

• More

Machine Learning

EPILEPTIC SEIZURE PREDICTION

This project demonstrates the classification of EEG (Electroencephalogram) data using various machine learning algorithms from scikit-learn library. The EEG data contains readings from different patients, and the project classifies whether a patient is epileptic or not based on the EEG readings with an accuracy of 97%.

• More

REACT APP

PROJECT MANAGEMENT APP

This project is a simple project management application built using React. It allows users to create projects, add tasks to those projects, and manage them effectively.

• More

Publications

biomedical signal processing and control

Time series signal forecasting using artificial neural networks: An application on ECG signal

Certifications

Machine Learning Engineer Nanodegree

Infosys Certified Software Programmer

Data Science and Machine Learning Track- 30 days of google cloud challenge

T-REMUS MISSION SIMULATOR TO COLLECT DATA AT RIVER PLUMES FROM REGIONAL OCEAN MODELING SYSTEM
(ROMS)

Background

The vast and dynamic nature of the world's oceans has long captivated the scientific community due to the intricate web of complexities it presents. Among the many puzzles that oceanographers seek to unravel, the behavior of water at river plumes stands as a key enigma with far-reaching implications. Understanding water mixing at river plumes with the ocean is vital for comprehending ecological, climatic, coastal, and economic impacts (3)(4). This understanding includes effects on marine ecosystems, climate patterns, coastal processes, pollution dispersion, fisheries, and disaster response. River plumes form as a result of the merging of river water into the ocean, and the densities and particulate compositions of the river and ocean waters differ significantly, leading to a visible contrast between the river plume and the surrounding oceanic water.Collecting ocean data, in general, is an arduous task. Moreover, gathering data at river plumes presents its own unique challenges due to the distinctive dynamics of water in such areas. The scarcity of sufficient samples has proven to be the biggest hindrance in enhancing the understanding of the ocean. Over time, ocean data collection methods have undergone significant evolution. This transition has moved from manual boat-based (6) methods to utilizing advanced Unmanned Underwater Vehicles (UUVs) equipped with various sensors (7),(8),(9).In recent times, the data collection methods have improved profoundly. The use of Unmanned Underwater Vehicles (UUVs) has begun for data collection and exploratory purposes. There are times when UUVs were used to study the river plumes (7),(8),(9) but all the missions had limitations of reduced autonomy. T-REMUS represents one such vehicle that is deployed on missions to collect high-resolution ocean data from specific locations. However, the vehicle continues to run on preset missions and it lacks position correction abilities while under water. The current work focuses on developing a tool that will enable conduction of virtual missions allowing safe and effective ways of testing and training the UUVs.

Problem

The research aims to address the limitations of testing and training of an Unmanned Underwater Vehicle (T-REMUS) in a real ocean environment which in turn can be used to deal with the difficulties of collecting ocean data, specifically at river plumes.

Technical Discussion

The oceanographers seek to study the fundamental processes governing the mixing in the ocean, notably shear stratified and convective turbulence, and a considerable portion of this occurs within the plume front. Presently, the understanding of mixing evolve over time and space within this region remains incomplete, and the dynamics of water exchange between the plume source, plume body, and the front remain enigmatic. This complexity arises from the formidable challenge of accurately measuring and predicting turbulence in the confined, swiftly moving frontal area. Traditional methods, such as moorings, offer only sporadic data snapshots, while vertical measurements conducted from shipboard instrumentation are compromised by strong currents, making it difficult to maintain a fixed position. Moreover, conventional sampling approaches employed by Autonomous Underwater Vehicles (AUVs) are constrained by predetermined mission routes or gateway buoy tracking systems, limiting their ability to effectively track and investigate the dynamic nature of a front. Consequently, the situation calls for an innovative and unconventional approach to overcome these limitations and shed light on these intricate processes.

T-REMUS (1) is a modified version of the standard Remote Environmental Monitoring System (REMUS) originally built by Hydroid Inc. The vehicle allows downloading of the collected data post it’s mission by interfacing with a Windows laptop computer. It has additional sensors attached to it which enable the vehicle to collect oceanographic data such as pressure, temperature, salinity, bathymetry, water velocity, and shear. T-REMUS also measures the gradients of pressure and temperature. The robot's built-in navigation system relies on various tools like Long Baseline (LBL) and Ultra-Short Baseline (USBL) acoustic devices, a Global Positioning System (GPS), data from an Acoustic Doppler Current Profiler (ADCP), and its own compass.

FIGURE 1: T-REMUS

T-REMUS (1) offers a promising means of data collection, utilizing its capabilities to navigate through the ocean depths and gather critical information. However, the limitations of T-REMUS, especially in autonomously determining its position while submerged, highlight the pressing need for further advancements in this domain. While the vehicle can establish its geographic location when on the water's surface, this ability diminishes when submerged. Consequently, the vehicle relies on pre-positioned gateway buoys to assess its proximity to fixed reference points, shaping its subsequent course of action. The vehicle's awareness of its precise position is essential due to the potential misguidance caused by ocean waves, necessitating autonomous correction capabilities. This underscores the need for enhanced autonomy in the vehicle's operations. However, training and testing actual Unmanned Underwater Vehicles (UUVs) in real ocean conditions to achieve autonomy is both inefficient and uneconomical. Hence, there arises a demand for an innovative approach to transcend this reliance and amplify the efficiency of oceanic data collection. Specifically, this research concentrates on the data collection of ocean at the river plumes.

FIGURE 2: View from an early stage plume leaving the Merrimack River mouth (5).

The virtual ocean will be built using Regional Ocean Modeling System (ROMS)(2). ROMS is a numerical model used in oceanography and coastal research to simulate and study the physical and biogeochemical processes of ocean and coastal environments. ROMS is a widely used tool for understanding ocean circulation, water temperature, salinity, and other oceanographic phenomena. The output of the model consists of several dimensions, co-ordinates, variables, attributes. The dimensions of the data describes the location, timestamp for the variable data. The variables can be many such as salinity, pressure, north or east velocity of water, temperature, etc. This can then be used as ocean waters to perform virtual exploratory missions and collect data at specific locations.

FIGURE 3: Horizontal profile of Salinity changes with respect to time at river plume plotted from ROMS output data

Approach

This research proposes a solution that intersects technology, modeling, and autonomy to bridge the gap between T-REMUS' current capabilities and the imperative for efficient, independent data collection. The problem advocates for the development of a sophisticated simulator, intricately designed to emulate TREMUS’ behaviors within a virtual ocean environment. This environment would be meticulously constructed using the output of the Regional Ocean Modeling System (ROMS), which provides a digital platform to replicate real-world ocean scenarios.

References

1. http://www.smast.umassd.edu/Turbulence/remus.php
2. https://www.myroms.org/wiki/RegionalOceanModelingSystem(ROMS)
3. Levine, E.R., Goodman, L., & O'Donnell, J. (2009). Turbulence in coastal fronts near the mouths of Block Island and Long Island Sounds. Journal of Marine Systems, 78(3).
4. MacDonald, D.G., Carlson, J.O., & Goodman, L. (2013). On the heterogeneity of shear-stratified turbulence: Observations from a near-field river plume. Journal of Geophysical Research: Oceans, 118, 6223-6237.
5. Figure from Piffer-Braga, 2023, in prep.
6. Spicer, P., Huguenard, K., Cole, K. L., MacDonald, D. G., & Whitney, M. M. (2022). Evolving interior mixing regimes in a tidal river plume. Geophysical Research Letters, 49, e2022GL099633.
7. Delatolas, N., MacDonald, D. G., Goodman, L., Whitney, M., Huguenard, K., & Cole, K. (2023). Comparison of structure and turbulent mixing between lateral and leading-edge river plume fronts: Microstructure observations from a T-REMUS AUV. Estuarine, Coastal and Shelf Science, 283, 108234. ISSN 0272-7714.
8. Zhang, Y., et al. (2022). Autonomous Tracking of Salinity-Intrusion Fronts by a Long-Range Autonomous Underwater Vehicle. IEEE Journal of Oceanic Engineering, 47(4), 950-958.
9. Fisher, A. W., Nidzieko, N. J., Scully, M. E., Chant, R. J., Hunter, E. J., & Mazzini, P. L. F. (2018). Turbulent mixing in a far-field plume during the transition to upwelling conditions: Microstructure observations from an AUV. Geophysical Research Letters, 45, 9765–9773.

• Back

Contact