Thribhuvan Rapolu
About me
I am a graduate student at UT Austin working on Computer Vision and Stochastic Reinforcement Learning, advised by Dr. Chandrajit Bajaj. My research focuses on using stochastic reinforcement learning for time-dependent vision tasks, replacing deep learning where data is limited. I have a strong research background around pose estimation domain and stay up to date in this field. Along with my research, I have industry experience from internships at Dell during my undergrad studies. While interested in GPU optimization, I am still building hands-on experience. I am currently seeking internship opportunities to grow my skills and make an impact.
Education
MS in Computer Science
School/College: The University of Texas at Austin
Duration: Aug 2025 – May 2027
Relevant Coursework: Predictive Machine Learning (Prof. Chandrajit Bajaj), Generative Visual Computing (Prof. Qixing Huang)
B.Tech in Computer Science
School/College: Mahindra University
CGPA: 8.96/10
Duration: Jul 2021 – May 2025
Relevant Coursework: High Performance Computing, Reinforcement Learning, Machine Learning, Control Systems
Current Research
Computer Vision Machine Learning Stochastic Reinforcement Learning- Working on a research paper "Hamiltonian Non-Parametric Belief Propagation For Pose Estimation And Tracking" under Prof. Chandrajit Bajaj
- Implementing Differential Non Parametric Belief Propagation (DNBP) for tracking humans in 3D, extending prior applications from simple pendulum and spider environments to complex human motion
- Currently developing Stochastic RL solution as part of ongoing research
- Helping Professor on a book "Learning Hamiltonians, Through Manifold Optimization and Generalized Flows"
Work Experience
University of Texas at Austin
Research Assistant (Prof. Chandrajit Bajaj) | Mar 2024 – Present
- Working on a research paper "Hamiltonian Non-Parametric Belief Propagation For Pose Estimation And Tracking"
- Helping Professor on a book "Learning Hamiltonians, Through Manifold Optimization and Generalized Flows"
- Implementing Differential Non Parametric Belief Propagation (DNBP) for human pose estimation and tracking
Winter Intern (Manufacturing Team) | Feb 2025 – May 2025
- Built a PLC Dashboard in Power BI, processing 1M+ trace logs into insights and integrating AI for advanced analytics
- Developed an Angular + .NET application for order preprocessing with unified backend–frontend deployment on KOB
- Supported the Production Planning Engine with Pytest automation and data visualizations (plotly)
Project Intern (DevOps Support Team) | Jun 2024 - Aug 2024
- Implemented semantic-based routing and text-image responses chatbot using RAG for entire Dell Confluence Docs
- Built CI/CD pipeline for MongoDB. Explored fine-tuning LLM models as POC
- Tools & technologies used: Python, Pandas, Langchain, Postgres, MongoDB, GitLab, CI/CD
FaceIntel Technologies
Intern | Jan 2024 - Mar 2024
- Tools & technologies used: Python, Pandas, HTML, CSS, Javascript
- Analyzed data from Excel files to extract meaningful insights, identifying key trends and patterns and used them to create a detailed web dashboard.
Tools
Frameworks/Tools: Stable-Baselines3, OpenAI Gym, PyTorch, Jax, Scikit-Learn, Tensorflow, Plotly, Matplotlib, NetworkX, Beautiful Soup, Selenium, Pandas, Numpy, Langchain, OpenCV, Flask, CUDA C++, Git, Docker, SQL, Android Studio, UNIX
Certifications: Machine Learning A-Z, Tensorflow Developer, Sequence Models, Convolution Neural Network
Blogs
These are my notes that I wrote for summarizing papers and code.
- PHALP-4D Humans WorkFlow
- HMR2.0 with code
- 4D Humans Paper
- 4D Humans Code
- DNBP Simplified
- Multi Term Adam Explained
- DNBP Loss Backpropagation
- Survey around 3D Human Pose estimation and tracking
Projects
Computer Vision Deep Learning Transformers- For the Amazon ML Hackathon, trained a Donut Transformer on a custom image dataset
- Automatically identifies product entities and their values without specifying entity type
- Implemented end-to-end document understanding pipeline for product information extraction
Reinforcement Learning Control Systems DDPG- Explored optimal control in low-bandwidth settings by building estimators for signal recovery and real-time scheduling
- Used DDPG to address challenges in data reconstruction and transmission
- Developed robust control policies for communication-constrained environments
NLP RAG LLM- Developed an RAG-based LLM model using pre-trained Llama-2 model with Langchain
- Applied to entire NVIDIA documentation gathered using Beautiful Soup web scraping techniques
- Created intelligent assistant for NVIDIA SDK and toolkit learning
Machine Learning Graph Theory Network Analysis- Identified and grouped nodes within graphs through machine learning that share common structural characteristics
- Revealed clusters of individuals with similar traits or interests
- Applied graph neural networks for community detection and network analysis
Android Development Kotlin Mobile App- Countdown timer app that pauses video/audio playback and locks screen when timer stops
- Developed in Kotlin with focus on user experience and system integration
- Implemented media control and system lock functionality
Computer Vision Desktop App OpenCV- Tools & technologies used: Python, OpenCV, dlib, PushOver API
- Notifies user to take a break while working if time limit exceeds for more than 20 minutes and many more features like screen distance and drowsiness detection
Web Application ChatGPT API Integration- Tools & technologies used: Javascript, Openstreetmap API
- A ChatGPT based web application that helps the user with their tour plan
- Users can interact with ChatGPT to inquire about tour plans, and it will provide suggestions for locations on a map
Discord Bot Automation Python- Tools & technologies used: Pandas, Hikari-Lightbulb, Google API
- A bot for my university that automatically accesses the mess menu file and pings everyone in the server 30 minutes before breakfast, lunch, snacks, and dinner to notify them of the meal options
Discord Bot Authentication University Integration- Tools & technologies used: Pycord, PHP, SQL, Python
- Configuring Discord as a communication platform for a university, ensuring that only authorized individuals can access course-specific Discord servers
File Manager Automation Python- Tools & technologies used: Watchdog
- Automatically moves files downloaded from downloads folder to Pictures or Videos or Documents folder based on file type
Achievements
- Scholarship of $1,000 for exceptional academic performance by Mahindra University. Oct 2023 & Nov 2024
- The Mirai CTF: Achieved 2nd place in a Capture The Flag (CTF) Cybersecurity challenge. Apr 2024
- Data Challenge: Secured 3rd place for implementing an ML solution on a provided dataset. Apr 2024
- Dell Hackathon: 2nd Winners for working on intimate user to take a break while working. Oct 2023
- The Mirai CTF: 3rd place in a Capture The Flag (CTF) Cybersecurity challenge. Apr 2023
- Data Challenge: Secured 3rd place for implementing an ML solution on a provided dataset. Apr 2023
- The 418 Hackathon: 1st place for integrating ChatGPT with a location-based API. Apr 2023
- Technomania: 2nd place in challenges based on Competitive Coding (DSA). Feb 2023
Contact