ABOUT ME

Hello, I'm Harmesh,
a college student with
a strong interest
in technology and
a problem-solving mindset

Profile picture

See small bits of my project works on Github, quick thoughts on Twitter, and a full resume on LinkedIn.

What I'm doing

  • Machine Learning Icon

    Machine Learning

    Building intelligent systems with modern techniques and clean designs.

  • Data Science Icon

    Data Science

    Extracting insights from data using statistical and analytical techniques.

  • Web Development Icon

    Web Development

    Designing and developing responsive, dynamic websites and apps.

Skills

Turning ideas into reality with code and creativity

Machine Learning
Python Python
Pandas Pandas
Numpy Numpy
MySQL MySQL
Oracle Oracle
Matplotlib Matplotlib
Tableau Tableau
Frontend Development
HTML HTML
CSS CSS
React React
Backend Development
Java Java
FastAPI FastAPI
Flask Flask
DevOps and Tools
Docker Docker
Git Git
GitHub GitHub
Jenkins Jenkins
Linux Linux (Ubuntu)
Projects

Building solutions to real-world problems, one idea at a time.

  • Grocery Basket Project
    Web HealthCare
    Technologies Used: PyTorch, OpenCV, TensorFlow, BeautifulSoup, Hugging Face
    Duration: Aug’ 24 - Oct’ 24
    Project Highlights:
    Developed a ResNet-based deep learning model for skin disease classification, achieving 92% accuracy across 10+ conditions.
    Implemented real-time multi-class classification for AI-driven diagnostic recommendations, processing 50+ images/min.
    Automated web scraping of 500+ disease records from Mayo Clinic, including disease causes, effects, and prevention measures.
    Integrated RAG model for symptom-to-disease matching, achieving 98% precision in providing causes, effects, and prevention.
  • Grocery Basket Project
    Video Classification Project
    Technologies Used: PyTorch, YOLOv8, LSTM, CUDA, GNNs (PyG)
    Duration: Feb 25 - Present
    Project Highlights:
    Achieved real-time 30+ FPS pose extraction by optimizing YOLOv8 with Torch-TensorRT and FP16 mixed precision.
    Implemented multi-person key point tracking (17 COCO points/frame) with IoU-based ID matching for temporal consistency.
    Built real-time multi-person interaction analysis (Euclidean/motion + YOLO weapon detection) with 98% pose accuracy.
    Optimized CUDA memory with automatic cache management and cuDNN benchmarking for sustained high throughput.
    Transformed pose sequences into temporal graph data (PyG) for GNN-based violence detection over 5-frame sliding windows.
Experience

Real-world challenges taught me what code alone couldn’t.

School Icon

Software Development Intern

Cloud It Solutions
Jun’ 22 – Jul’ 22
Built a Generative Adversarial Network (GAN) to generate realistic 28x28 images using adversarial training

School Icon

Freelance Frontend Developer

Six Square Builders
Nov’ 24 - Feb’ 25
Designed and developed a full-scale business website from scratch using JavaScript and React.js, enhancing brand presence.

Education

Rooted in strong fundamentals, growing through real-world application.

School Icon

B.Tech Computer Science

Aug 2022 - Aug 2026 (Expected)
Specialization: Machine Learning & Data Science
Lovely Professional University, Punjab

School Icon

Higher Secondary

April 2021 - March 2022
Sri Chiatanya, Chennai.

School Icon

Higher Secondary

April 2021 - March 2022
Sri Chiatanya, Chennai.

Certifications

Building solutions to real-world problems, one idea at a time.

img

Devops and Software Development

Docker • Jenkins • Cloud • CI/CD

img

Machine Learning and Data Science

Python • EDA • Deep Learning

img

Artificial Intelligence

Machine Learning • Generative AI

Achievements

Rooted in strong fundamentals, growing through real-world application.

Skin Disease Prediction Hackathon

TabuQuest AI/ML Hackathon

February 9-10, 2025
Built an LLM-powered system for smart filtering and answering from tabular datasets at TabuQuest, an AI/ML hackathon, focusing on reducing data load and enhancing answer precision from CSV files.
Data Understanding: Implemented column-name semantic comparison to understand table structure without hardcoding.
Row Filtering: Designed an NLP-driven filtering pipeline using three vector strategies — date-based, ID-based, and keyword-based — to extract relevant rows.
RAG Pipeline: Developed a Retrieve-and-Generate model to answer natural language queries from filtered CSV data with 80–90% accuracy.
Skin Disease Prediction Hackathon

Skin Disease Prediction Hackathon

August 28-29, 2024
Built a skin disease prediction system and a symptom-based LLM during an open-topic hackathon by GeeksforGeeks, enabling users to input symptoms and receive disease details, effects, precautions, and medical domain recommendations.
Data Extraction: Scraped data from 500+ diseases using BeautifulSoup for symptom, cause, effect, and prevention details.
Symptom Prediction: Designed a RAG-based model for accurate symptom-to-disease mapping.
Skin Disease Classification: Trained a ResNet-based model achieving high-accuracy skin disease detection from images.
Skin Disease Prediction Hackathon

Skin Disease Prediction Hackathon

August 28-29, 2024
Built a skin disease prediction system and a symptom-based LLM during an open-topic hackathon by GeeksforGeeks, enabling users to input symptoms and receive disease details, effects, precautions, and medical domain recommendations.
Data Extraction: Scraped data from 500+ diseases using BeautifulSoup for symptom, cause, effect, and prevention details.
Symptom Prediction: Designed a RAG-based model for accurate symptom-to-disease mapping.
Skin Disease Classification: Trained a ResNet-based model achieving high-accuracy skin disease detection from images.
Get in touch

For work inquiries
please email:
harmeshgopinathan@gmail.com

More links: Github, Twitter, and Linkedin.