Hi, I'm

Poorvi Bellur

Computer Science Student Β· Data Science Minor

I'm a Computer Science undergrad at RVCE, Bangalore (Data Science minor), building ML systems aimed at real-world problems β€” from a multi-modal highlight-detection pipeline built during my internship at Samsung R&D, to a published safety-aware reinforcement learning framework for lunar rover navigation. Graduating in 2027.

Poorvi Bellur

About

A little about me

I care about solving real-world problems through technology, not just building things that are technically interesting. That's shown up in a published, safety-aware reinforcement learning framework for lunar rover navigation; a multi-modal ML pipeline I built during my internship at Samsung R&D; Archivex, a fully local, agentic RAG assistant that answers technical questions straight from PDF documents; IoT-based systems like BruxSense, a wearable bruxism monitor I presented at Maker Faire Hannover, and a predictive model for reducing human-elephant conflict in Tamil Nadu's Valparai region; and, on the culture side, a searchable database of Indian heritage monuments and VARA, an e-commerce platform for Channapatna's traditional wood-artisans. Machine learning and data engineering, including distributed hyperparameter optimization on Apache Spark, tie a lot of it together.

Outside of coursework, I'm Treasurer of Alaap, RVCE's music club, and play in our band Ameya β€” we've won multiple Battle of Bands competitions, including at BITS Goa, BMSCE, and the IIT-Bombay Bangalore elims. I also placed 2nd at the Craft n Code hackathon at IIIT Bhubaneswar.

Based in Bangalore, India
Studying B.E. Computer Science & Engineering
Minor Data Science
Graduating 2027

Education

Academic background

RV College of Engineering (RVCE), Bangalore

2023 β€” 2027

B.E. in Computer Science & Engineering Β· Minor in Data Science

CGPA: 9.59 / 10 (through 6th semester).

School / Pre-Univ

2021 β€” 2023

BASE PU College Β· Karnataka PU Board

Scored 98.3% in Class 12 (State Board); ranked 7th in Karnataka PU Board, Class of 2023.

Skills

Tools & technologies

Languages

Python C Java JavaScript SQL

Generative AI & Agentic Systems

RAG Pipelines ChromaDB Ollama Llama 3 Mistral Phi-3 HuggingFace Transformers BLIP-2 Sentence-Transformers

Data Science & ML

Pandas NumPy scikit-learn XGBoost PySpark TransNetV2 CLIP YAMNet CLAP Matplotlib

Web & Software

React FastAPI Express Node.js Django MongoDB Firebase Git GitHub Postman UiPath Studio

Tools & Platforms

Apache Spark Docker Render GCP AWS

Projects

Things I've built

This section builds itself: every card below is fetched live from github.com/poorvi4567, and Claude reads each repo's README to write the summary and tags β€” nothing here is hand-typed. The ✨ badge marks projects Claude has already summarized; the rest show the repo's own GitHub description until that happens.

Loading projects from GitHub…

Experience

Internships & involvement

Samsung R&D

Jun 2025 β€” Dec 2025

Virtual Internship

View certificate
  • Designed and shipped an end-to-end multi-modal ML pipeline β€” TransNetV2 and CLIP for video/scene understanding, YAMNet and CLAP for audio embedding β€” to automatically detect key highlight moments in slow-motion and hyperlapse sports videos.
  • Embedded high-quality voice/audio into auto-detected highlight moments, streamlining the sports-video production workflow end to end.

ALAAP, Music Club of RVCE

Jun 2026 β€” Present

Treasurer

  • Led the club's band, Ameya, to wins across multiple Battle of Bands competitions, including BITS Goa, BMSCE, and the IIT-Bombay Bangalore elims.

Highlights

Research & recognition

A published paper, a wearable device shown internationally, and a hackathon win β€” three projects worth a closer look.

IEEE Access Β· Vol. 14, 2026

Safety-Aware Reinforcement Learning for Lunar Rover Navigation Using Orbital DEMs

Co-authored with a research team at RVCE, this paper presents a digital twin for training lunar rover navigation policies on real NASA orbital elevation data. A reinforcement learning policy (PPO) is paired with a vision-based hazard-detection module and a reactive safety filter, trained through a five-axis curriculum that gradually increases terrain difficulty. Across 20,410 evaluation episodes, the system reached 96%/85% success on moderate/high-complexity terrain β€” versus 15% for heuristic baselines β€” while cutting collision rates from 35% to 10%.

Read the paper

Maker Faire Hannover Β· Aug 2026

BruxSense β€” Wearable Bruxism Monitor

Presented BruxSense, a wearable device that detects nocturnal bruxism (teeth grinding and jaw clenching), at Maker Faire Hannover β€” turning EMG and jaw-motion sensor data into a real-time monitoring dashboard clinicians can use to track patients overnight.

View on GitHub

2nd Place Β· Craft n Code Hackathon

YouTube Thumbnail Generator

Built an application that automatically generates optimized YouTube thumbnails from video content, placing 2nd at the Craft n Code hackathon at IIIT Bhubaneswar.

View on GitHub

Achievements

Certifications & awards

AWS Cloud Practitioner Essentials

Amazon Web Services

View certificate

Craft n Code Hackathon

2nd Place Β· IIIT Bhubaneswar Β· Nov 2024

KCET Rank 456

Karnataka Common Entrance Test Β· 2023

Data Science for Engineers

NPTEL

IEEE Access Publication

Safety-Aware RL for Lunar Rover Navigation

View paper

Contact

Let's talk

Open to internships, research collaborations, and interesting projects. Reach out below or find me on GitHub / LinkedIn.