Arjun Desur

Machine Learning Engineer & Researcher

Arjun Desur

I'm a Software/Machine Learning Engineer and researcher with a degree in Electronics and Communication Engineering from NIT Warangal. My work spans applied AI, computer vision, and intelligent agents, with experience building machine learning systems for real-world deployment. I have authored two IEEE-published research papers in deep learning and image processing and enjoy turning research ideas into practical products.

US Citizen · New York 3× IEEE Published Graduated 2026 Open to Full-time & Internships
Get in touch Resume View projects

About

I'm a graduate in Electronics and Communication Engineering from NIT Warangal with a deep interest in machine learning, artificial intelligence, and software engineering. I've interned at two of India's premier technical institutions, IIT Madras and NIT Warangal, where I worked on computer vision and deep learning research that was published at IEEE conferences.


My recent work spans LLM-powered autonomous agents, fraud detection on AWS, and RAG pipelines for document retrieval. I enjoy taking projects from research ideas to deployed systems and am particularly drawn to problems at the intersection of AI and real-world engineering.


Beyond AI, I am interested in quantitative finance, algorithmic trading, and financial markets. I enjoy exploring how machine learning, statistics, and data-driven decision making can be applied to trading strategies, market analysis, and fintech applications.

Location New York, USA / Bangalore, India (US Citizen)
Email arjungdesur@gmail.com
LinkedIn linkedin.com/arjundesur
Open to Full-time roles, Internships (ML / AI / SWE)

Education

Academic background

B.Tech in Electronics and Communication Engineering
National Institute of Technology, Warangal
2022 – 2026  ·  Warangal, India

Relevant Coursework: Machine Learning, Data Structures & Algorithms, Computer Architecture, Operating Systems, Data Networks, Cryptography, Probability Theory & Stochastic Processes, Microprocessors

Skills

Technologies & tools

Languages
Python SQL C++ HTML / CSS
ML & AI
PyTorch Scikit-learn XGBoost HuggingFace Transformers LangChain FAISS Ollama OpenCV
Frameworks & Tools
FastAPI Flask Streamlit REST APIs Git Jupyter NumPy Pandas
Cloud & DevOps
AWS SageMaker AWS EC2 AWS S3 Docker Linux

Experience

Where I've worked

May – July 2025
Machine Learning Summer Intern
NIT Warangal — National Institute of Technology, Warangal
Warangal, India
  • Developed a deep learning framework for military aircraft classification by integrating ResNet-50 and EfficientNet with multi-transformation feature fusion (PCA, KPCA, SVD, ICA, LDA) and XGBoost, achieving 99.33% accuracy on the MTARSI dataset and outperforming prior CNN-based approaches.
Published · IEEE ICONAT 2025
May – July 2024
Machine Learning Summer Intern
IIT Madras — Indian Institute of Technology, Madras
Chennai, India · TrityAEye / ITS Lab
  • Built a computer vision pipeline to filter ML outputs by vehicle category, automate frame extraction from 100+ surveillance videos, and annotate 2,000+ vehicle images across multiple truck classes using LabelMe.
  • Validated object detection outputs by cross-referencing tracking data against ground truth footage, supporting a model achieving ≥95% classification accuracy.
Aug – Oct 2023
Machine Learning Researcher
Digital Signal Processing Lab, NIT Warangal
Warangal, India
  • Proposed a hybrid Glowworm Swarm Optimization and K-Means clustering algorithm (KM-GSO) for color image quantization, achieving PSNR 26.10 dB / SSIM 0.92 vs. 14.88 dB / 0.17 for standard K-Means (K=64), outperforming 5 benchmark methods on the Kodak dataset.
  • Applied algorithm to Chandrayaan-3 mission images, achieving SSIM of 0.96.
Published · IEEE ICPC2T 2024

Projects

Things I've built

Autonomous AI Agent (ReAct)
Mar 2026

LLM-powered autonomous agent using the ReAct loop with LangChain, dynamically routing across web search (Tavily), Python execution, and document retrieval. Evaluated on 150+ multi-domain prompts at 380ms avg latency via Groq API (Llama 3).

Python LangChain Groq API ReAct
Fraud Detection on AWS
Feb 2026

Trained XGBoost and Random Cut Forest on 200,000+ imbalanced transaction records (492 fraud cases). Applied SMOTE + threshold tuning; achieved 92% recall on fraud class. Deployed via SageMaker endpoints with a REST API for real-time scoring.

Python XGBoost AWS SageMaker REST API
RAG Document Q&A Chatbot
Oct 2025

End-to-end RAG pipeline over 50+ PDFs using LangChain, FAISS, and Ollama (Llama 3) with 512-token recursive chunking and cosine similarity retrieval. Streamlit UI with sub-3s latency across 200+ test queries — fully offline.

Python LangChain FAISS Ollama Streamlit
Loan Default Prediction
Jun 2025

Machine learning model to predict loan default likelihood using historical applicant data. Performed extensive data preprocessing and EDA to handle missing values and class imbalance, applying SMOTE oversampling. Trained a Decision Tree Classifier for interpretable and actionable risk assessments for financial institutions.

Python Decision Tree SMOTE EDA Scikit-learn
Algorithmic Portfolio Optimization
Jan 2025

Implemented a Sharpe Ratio-based optimization model on 7 years of historical stock data (2016–2022) for AAPL, MSFT, GOOGL, TSLA, achieving 18% annualized return at 12% risk (Sharpe = 1.5). Built pipeline covering data collection, preprocessing, EDA, and portfolio optimization.

Python Pandas NumPy SciPy Matplotlib yfinance

Publications

IEEE Xplore

Enhanced Military Aircraft Classification Using ResNet and EfficientNet with Multi-Transformation Feature Fusion and XGBoost
IEEE ICONAT 2025 · Cosponsored by IEEE Bombay Section · September 2025
View on IEEE Xplore →
Enhanced Color Image Quantization Using Hybrid Glowworm Swarm Optimization and K-Means Clustering
2024 Third International Conference on Power, Control and Computing Technologies (ICPC2T) · January 2024
200+ views · 2 citations
View on IEEE Xplore →
Integrated Wearable Smart Band and Autonomous Mobile Robot for Elderly Care
2025 17th International Conference on Knowledge and Smart Technology (KST) · February–March 2025 · Bangkok, Thailand
View on IEEE Xplore →

Research

Research interests

My research focuses on deep learning, computer vision, and intelligent systems. I have worked on image classification using ensemble CNN architectures and feature fusion techniques, color image quantization with bio-inspired optimization, and autonomous robotic systems for assistive applications. These projects have produced three peer-reviewed publications in IEEE conferences. I am broadly interested in the application of machine learning to real-world engineering problems, including signal and image processing, autonomous agents, and intelligent decision-making systems.

IEEE Profile Google Scholar

Contact

Let's connect

Open to full-time ML/AI engineering roles and research opportunities. Feel free to reach out.

arjungdesur@gmail.com LinkedIn GitHub