Skip to content
View Mukesh-2005's full-sized avatar

Block or report Mukesh-2005

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Mukesh-2005/README.md

🎯 About Me

Mukesh = {
    "role"         : "Data Scientist & ML Engineer",
    "location"     : "Tenkasi, Tamil Nadu 🇮🇳",
    "education"    : "B.Sc Data Science @ MSU (2023–2026)",
    "email"        : "starmukesh2005@gmail.com",
    "linkedin"     : "linkedin.com/in/mukesh-k-6899902b7/",
    "github"       : "github.com/Mukesh-2005",
    
    "expertise"    : [
        "Object Detection (YOLO)",
        "NLP / Text Classification",
        "Ensemble Learning",
        "Feature Engineering",
        "Data Science Pipeline",
        "REST APIs"
    ],
    
    "learning"     : ["Computer Vision", "Transformers / LLMs", "PyTorch", "Production ML"],
    "open_to"      : ["Full-time Data Science", "ML Internships", "Collaborations"],
}

🏆 Highlights

  • 🤖 Traffic Vehicle DetectionmAP50: 0.749 | Real-time inference (4.2ms) | 6.9x improvement
  • 🎫 Automated Ticket Routing System — TF-IDF + Logistic Regression NLP classifier | 80% test accuracy | production-style confidence triage
  • 🎓 IBM Professional Data Science Certificate — 12-course specialization (Coursera) ✅
  • 🎯 End-to-End Projects — Data prep → Model training → Production deployment

🚀 Featured Projects

🚗 Traffic Vehicle Detection ⭐ FEATURED

Real-time Object Detection for Indian Traffic

  • Model: YOLOv8s
  • mAP50: 0.749 ✓
  • Inference: 4.2ms (real-time!)
  • Dataset: 1,300 perfectly balanced images
  • Achievement: 6.9x improvement by fixing data imbalance (6.2:1 → 1:1)

Performance by Class:

  • Car: 0.762 ✓
  • Bus: 0.880 ⭐ (Best)
  • Truck: 0.634
  • Motorcycle: 0.677
  • Auto-rickshaw: 0.795

Key Learning: Data quality > Model complexity

Tech: YOLOv8 | PyTorch | Python | Colab

🔗 GitHub Repo | Demo Video | Writeup

🎫 Automated Ticket Routing System ⭐ FEATURED

NLP Classifier — Production-Style Triage

  • Model: TF-IDF + Logistic Regression
  • Test Accuracy: 80% | Macro F1: 0.79
  • Classes: 4 departments (Billing, Technical, HR, General)
  • Diagnosed Failure: Fixed an HR-class 0% recall bug by identifying vocabulary overlap and rewriting training data

Production Features:

  • Confidence-based "needs human review" threshold (0.60)
  • Keyword-based urgent/normal priority tagging
  • Interactive CLI demo for live ticket testing
  • Feedback loop scaffolding for future retraining

Key Learning: Model architecture was correct from the start — the real fix was diagnosing why a class had zero signal in the data, not just tuning hyperparameters

Tech: Scikit-learn | TF-IDF | Python | Pandas | Joblib

🔗 GitHub Repo

💬 Social Media API

Production-Ready REST API — 45+ Endpoints

  • Status: 🟢 Live & Running
  • URL: https://social-media-api-ispt.onrender.com/docs
  • Features: Posts, comments, followers, smart feeds, trends
  • Database: 5-table relational schema with complex SQL
  • Deployment: Render (production-grade, $0 cost)

Endpoints:

  • User management (CRUD)
  • Post creation & engagement
  • Follow system
  • Personalized feeds
  • Trending algorithms

Tech: FastAPI | SQLAlchemy | SQLite | JWT | Pydantic

🔗 GitHub Repo

🧬 Diabetes Risk Prediction

Medical Classification — 92.2% Accuracy

  • Accuracy: 92.2%
  • Classes: 7 risk categories
  • Dataset: 2,000+ medical records
  • Best Model Log-Loss: 0.427

Methodology:

  • Compared OvA, OvO, Multinomial
  • Stratified cross-validation
  • Feature scaling & encoding
  • Class imbalance handling

Tech: Scikit-learn | Pandas | NumPy

🔗 GitHub Repo

🔐 Task Manager API

Role-Based Access Control

  • Auth: JWT token-based
  • Security: Password hashing
  • Features: CRUD, permissions, filtering
  • Database: SQLite with ORM

Capabilities:

  • User authentication
  • Role-based access
  • Task management
  • Priority filtering
  • Date-based sorting

Tech: FastAPI | SQLAlchemy | JWT | Pydantic

🔗 GitHub Repo

📚 Coursera Data Science Labs

30+ Hands-On ML Projects

  • Topics: Supervised & Unsupervised Learning
  • Algorithms: Classification, Regression, Clustering
  • Metrics: Accuracy, Precision, Recall, F1, AUC-ROC, MSE, RMSE
  • Models: Decision Trees, SVM, KNN, Linear Regression, K-Means, PCA, DBSCAN

Coverage:

  • Data preprocessing
  • Feature scaling & selection
  • Model evaluation
  • Hyperparameter tuning
  • Cross-validation

Tech: Scikit-learn | Matplotlib | Seaborn | Pandas

🔗 GitHub Repo


🛠️ Tech Stack

🐍 Languages & ML Frameworks



⚙️ Backend & APIs



🗄️ Databases



🧰 Tools & Environment



📈 GitHub Stats

GitHub followers GitHub User's stars Profile Views


🎓 Certifications & Education

Certification Organization Status
🏅 IBM Professional Data Science Certificate IBM / Coursera ✅ Completed
🏅 Machine Learning with Python IBM / Coursera ✅ Completed
🏅 Databases & SQL for Data Science IBM / Coursera ✅ Completed
📚 B.Sc Data Science Manonmaniam Sundaranar University ✅ Completed

📚 Currently Learning

Area Focus
👁️ Computer Vision Advanced YOLO, OpenCV, Image Processing
🗣️ NLP Transformers, LLMs, RAG Systems
🧠 Deep Learning Advanced PyTorch, Neural Architecture Design
🐳 DevOps Docker, Kubernetes, CI/CD Pipelines

💡 Data Science Philosophy

Data Quality > Model Complexity

My core principles:

  • EDA First — Understand data before modeling
  • Fix Imbalance Early — Balance classes before upgrading models
  • Feature Engineering > Model Size — Smart features beat complex models
  • End-to-End Thinking — Production-ready from day one
  • Iterate & Validate — Cross-validation, not lucky splits

Real Example 1 — Traffic Detection:

  • Initial: 0.108 mAP50 (broken model)
  • Root Cause: 6.2:1 class imbalance, not model
  • Solution: Fixed data (1:1 balance)
  • Result: 0.749 mAP50 (6.9x improvement!)

Real Example 2 — Ticket Routing NLP:

  • Initial: HR class at 0% precision and recall
  • Root Cause: Generic training phrasing overlapping with other classes, not the model
  • Solution: Rewrote HR examples with domain-distinctive vocabulary
  • Result: HR precision 0% → 85.7%

🌐 Connect With Me

Platform Link
💼 LinkedIn linkedin.com/in/mukesh-k
💻 GitHub github.com/Mukesh-2005
📧 Email starmukesh2005@gmail.com
🔗 Live API Social Media API Demo
🚗 Traffic Detection GitHub Repo
🎫 Ticket Routing NLP GitHub Repo

📊 Project Quality Metrics

Project Type Performance Status
🚗 Traffic Vehicle Detection Computer Vision mAP50: 0.749 ✅ Production
🎫 Ticket Routing NLP NLP / Text Classification Accuracy: 80% ✅ Completed
💬 Social Media API Backend 45+ Endpoints 🟢 Live
🧬 Diabetes Prediction Classification Accuracy: 92.2% ✅ Completed
🔐 Task Manager API Backend JWT + Role-Based ✅ Completed
📚 Data Science Labs Educational 30+ Projects ✅ Completed

Pinned Loading

  1. Coursera-ML-Labs Coursera-ML-Labs Public

    My completed lab sessions from the Coursera Machine Learning course, with personal notes and enhancements.These notebooks represent my hands-on learning from structured labs. I didn’t build the alg…

    Jupyter Notebook 2

  2. spacex-classification-project- spacex-classification-project- Public

    🚀 Classification project using SpaceX launch data — part of Coursera’s Data Science Capstone. Includes data collection, wrangling, EDA, geospatial mapping, and dashboarding.

    Jupyter Notebook 1

  3. traffic-vehicle-detection traffic-vehicle-detection Public

    "Week 2: Traffic Vehicle Detection - YOLOv8s with balanced dataset (mAP50: 0.749)"

    Python

  4. Traffic-detection-app Traffic-detection-app Public

    Python

  5. Social-Media-API- Social-Media-API- Public

    Python