Portfolio - Yogeshwaran K

Yogeshwaran K

ML Engineer

Resume

2

Machine Learning Projects

2

Deep Learning Projects

About Me

I am a Computer Science Engineer passionate about building robust software, intelligent data-driven applications, and scalable systems. My background bridges core software engineering with modern applied artificial intelligence spanning database design, clean RESTful API architecture, predictive analytics, and deep learning solutions.
I focus on crafting clean, maintainable code and solving complex technical challenges from end to end. Whether architecting lightweight microservices with FastAPI, creating responsive user experiences with React and Streamlit, or optimizing computational pipelines for performance, I enjoy turning abstract technical concepts into reliable, production-ready tools.

Key Skills

Python

Machine Learning

GenAI

Deep Learning

My Projects

Life Insurance Cost Prediction

Domain/Function: FinTech

RAG-Based Telecom Support Bot

Domain/Function: Telecom Automation

Medicinal Leaf & Plant Classification

Domain/Function: AgriTech

Car Damage Classification

Domain/Function: Automobile Industry

Awards & Certificate

Python: Beginner to Advanced For Data Professionals

SQL for Data Science

Math and Statistics For AI, Data Science

Master Machine Learning for Data Science & AI: Beginner to Advanced

Let's Connect

Feel free to get in touch with me. I am always open to discussing new projects, creative ideas or opportunities to be part of your visions.

Download Resume

Resume

Mail Me

[email protected]

https://www.linkedin.com/in/yogeshwaran-ka/

Call Me

+91 6382437292

GitHub

https://github.com/yogeshwaran-cse

Machine Learning Project

Life Insurance Cost Prediction

Domain/Function: FinTech

Created an interactive web tool that estimates health and life insurance premiums based on individual lifestyle and health metrics. Delivers accurate, real-time pricing forecasts through an intuitive user dashboard.

View Project
Engagement

View Project
on GitHUB

Other Tools

Machine Learning

Regression Modeling

Streamlit

Scikit-learn

Pandas

The front End of the Life Insurance project using Streamlit

The Result of the Life Insurance prediction based on the given inputs that is given in the UI

Project Details

🧬 Engineered an end-to-end insurance premium forecasting pipeline, designing composite lifestyle and medical risk-scoring formulas to optimize policy pricing models across demographic and biometric factors.
📈 Trained and tuned an XGBoost Regressor via RandomizedSearchCV, outperforming baseline regression models to achieve a 0.9950 R² score and reducing test RMSE by 65.5% to 7623.97.
🚀 Deployed an interactive Streamlit application utilizing serialized pipelines, delivering real-time premium estimates with an average 3.55% MAPE and reducing high-margin prediction errors (>10%) down to 5.1%.

Machine Learning Project

RAG-Based Telecom Support Bot

Domain/Function: Telecom Automation

Designed an automated customer support assistant for telecom inquiries that accurately fetches plan details and FAQs. Provides verified, hallucination-free answers in real time through a responsive, low-latency web service.

View Project
on GitHUB

Other Tools

Generative AI

Langchain

Vector AI

FastAPI

UI of the Telecom Support system by using streamlit as a frontend

The resonse of the chat system after asking the relevent questions

Project Details

🤖 Architected an enterprise-grade telecom RAG chatbot integrating Gemini and ChromaDB across 3 heterogeneous databases, achieving a 100% source retrieval hit rate and an overall 91.4% RAG triad score.
🎯 Engineered LangChain prompt workflows with grounding filters, collaborating with domain experts to eliminate hallucinations, delivering 98.1% factual faithfulness and 100% negative constraint compliance on out-of-scope queries.
⚡ Optimized the backend deployment into a lightweight FastAPI serverless architecture, slashing package size by 96.8% to 18.5 MB while accelerating in-memory vector search latency to 42.8 ms.

Deep Learning Project

Medicinal Leaf & Plant Classification

Domain/Function: AgriTech

Built an accessible digital tool that identifies 80 medicinal plant species directly from leaf photographs. Runs instantly in the user's web browser without requiring external server processing or cloud fees.

View Project
Engagement

View Project
on GitHUB

Other Tools

Computer Vision

Deep Learning

The Frontend consist of upload section either choose or drag and drop option

The Inference Page after the Image is classified

Project Details

🌿 Engineered a botanical classification engine using MobileNetV2 transfer learning across 80 medicinal plant classes, achieving an 89.43% Top-1 accuracy and a 99.11% Top-5 categorical accuracy on unseen test evaluations.
🔍 Conducted detailed error analysis on test datasets to resolve morphological leaf misclassifications, achieving a 91.34% macro precision and securing a perfect 100% F1-score across 24 distinct plant species.
💻 Deployed an optimized TensorFlow.js WebGL pipeline into a full-stack React application, delivering client-side browser inference at 13.44 ms latency with zero backend cloud compute costs.

Deep Learning Project

Car Damage Classification

Domain/Function: Automobile Industry

Built an automated vehicle inspection tool that detects and classifies car exterior damage across 6 categories from photographs. Provides real-time claims evaluation and damage severity checks through an accessible interactive web application.

View Project
Engagement

View Project
on GitHUB

View Project
on

Other Tools

Pytorch

Deep Learning

ResNet

Streamlit UI for Car Damage Detection

Model Evaluation and visualize in Confusion Matrix

Project Details

🚗 Multi-Class Architecture: Built an end-to-end computer vision pipeline categorizing car damage into 6 distinct classes across 2,300 images, progressing through 5 model iterations from baseline CNNs to fine-tuned ResNet-50.
🎯 Bayesian Optimization: Tuned ResNet-50 hyperparameters via Optuna TPE search (learning rate: 3 x 1e-3, dropout: 0.473), achieving 81.57% validation accuracy with 0.80 macro F1-score across 575 test samples.
🚀 Interactive Deployment: Deployed an end-to-end interactive Streamlit web dashboard for real-time inference, featuring automated ImageNet preprocessing, instant photo preview, and human-readable damage diagnosis.

Already have an account? Sign in