🧠 Artificial Intelligence

Hi, I'm Harshvardhan

|

B.Tech in AI at IIT Gandhinagar — building intelligent systems at the intersection of deep learning, agentic AI, and MLOps.

View Projects Get In Touch
Scroll Down

01. About Me

I'm a second-year Artificial Intelligence undergraduate at IIT Gandhinagar, driven by a deep curiosity for how machines learn, reason, and act autonomously.

My work spans satellite imagery classification with CNNs, biomedical signal processing, hyperparameter optimization, and building end-to-end agentic AI systems. I love turning complex ML pipelines into deployable, real-world applications.

Education

🎓

B.Tech in Artificial Intelligence

IIT Gandhinagar

2024 — Present
📚

Class XII

RSML, Latur

2022 — 2024
🏫

Class X

Vrindavan English School, Aurad

2021 — 2022

02. Projects

A selection of things I've built — each one a step deeper into AI.

🛰️
Computer Vision

CNN-based Satellite Imagery Classification

Land-use classification from Sentinel-2 RGB satellite imagery over Delhi NCR using deep learning.

  • Trained a ResNet-18 model with GeoJSON-based spatial filtering and transfer learning to address class imbalance
  • Full pipeline from raw .tif raster processing and .npy feature extraction to geospatial visualization
  • Evaluated using Weighted F1-score and confusion matrix analysis
PyTorchResNet-18Transfer LearningGeoJSON
GitHub
🫁
Deep Learning

Breathing Anomaly Detection using 1D CNNs

Classifying apnea/hypopnea events from multimodal sleep signals using deep 1D convolutional networks.

  • Built a PyTorch pipeline with 1D SleepCNN across Nasal Airflow, Thoracic Movement, and SpO₂ signals from 5 participants
  • Applied Butterworth bandpass filtering and SpO₂ upsampling with sliding-window feature matrices (30s, 50% overlap)
  • Achieved >90% accuracy using Leave-One-Participant-Out Cross-Validation
PyTorch1D CNNSignal ProcessingLOPO CV
GitHub
🏠
MLOps

Dynamic House Rent Prediction Engine

ML-powered rent predictor across 4 cities with systematic hyperparameter optimization comparison.

  • Built a Random Forest model; compared Grid Search, Random Search, and Bayesian Optimization (Optuna) across 60 trials
  • Deployed via Streamlit UI with Docker containerization and Hugging Face Spaces hosting
  • Integrated LabelEncoder preprocessing and Trackio experiment tracking
Scikit-LearnOptunaStreamlitDocker
GitHub
🤖
Agentic AI

Autonomous Gmail Agent

AI-powered email manager with autonomous classification, task extraction, and smart reply generation.

  • Engineered with FastAPI + Google Gemini for email classification, task extraction, and Calendar event creation
  • Built priority scoring engine with confidence-based decision logging
  • Deployed on Streamlit with manual approval gates, Google OAuth 2.0, and SQLite persistence
FastAPIGeminiStreamlitOAuth 2.0
GitHub

03. Skills

💻

Languages

Python
🔧

Tools & Frameworks

PyTorch Scikit-Learn Pandas NumPy Matplotlib FastAPI Streamlit Docker Optuna LLM APIs Gmail API Google Calendar API
🧠

AI / ML Concepts

CNNs Transfer Learning Agentic AI LLM Integration MLOps Hyperparameter Optimization

04. Experience & Leadership

2026 — Present

Convener — Student Senate

IIT Gandhinagar

2024 — 2025

Curation Executive — TEDxIITGandhinagar

IIT Gandhinagar

2024 — 2025

Class Representative

IIT Gandhinagar

🏆 Award

Recipient of Explorer's Fellowship

IIT Gandhinagar

05. Get In Touch

I'm always open to collaborations, research discussions, or just a friendly chat about AI. Feel free to reach out!

✉️
harshvardhan.masalge@iitgn.ac.in
📱
+91 9665712875
github.com/harshmasalge
Say Hello 👋