Yash Zanwar

yashzanwar74@gmail.com

Pre-Final-year Integrated M.Tech CSE student at VIT, Vellore focusing on building reliable ML systems, MLOps pipelines, and generative AI tools. I enjoy solving practical problems, improving model reliability, and shipping reproducible ML. Currently learning: productionizing LLMs, advanced MLOps workflows, and scalable retrieval systems.

Projects

YouTube Comment Sentiment Analysis Chrome Extension

  • Built a Chrome extension to analyze YouTube video comments, offering sentiment classification (positive, neutral, negative), AI-based summaries, trend graphs, and word clouds.
  • Implemented the LightGBM model for sentiment classification, achieving an F-1 score of 87% on the testing data.
  • Managed experiments using MLflow and Dagshub, employing DVC for data/model versioning and automated training pipelines.
  • Built a CI/CD pipeline to containerize a ML model with a Flask API and deployed it using a rolling strategy on AWS Auto Scaling Groups.

Swiggy Delivery Time Prediction

  • Developed a delivery time prediction model using a Stacking Regressor on historical order, traffic, weather, restaurant prep-time, and GPS data.
  • Achieved MAE of 3.06 minutes and 15% improvement over baseline Linear Regression, with 85% R2 on test data.
  • Containerized the model and deployed via an automated CI/CD pipeline using AWS CodeDeploy and Auto Scaling Groups.

AI-Powered Equity Research and Q&A Tool

Technologies: Python, LangChain, Streamlit, OpenAI API, FAISS
  • Built an end-to-end Retrieval-Augmented Generation (RAG) system using OpenAI embeddings and FAISS to answer queries with cited sources.
  • Implemented full LangChain pipeline for ingestion, chunking, retrieval, and context-aware generation.
  • Developed an interactive Streamlit interface for seamless querying and visualization.

Coder Buddy: AI-Powered Application Builder

Technologies: Python, LangChain, LangGraph, Pydantic, LLMs (Groq Cloud API)
  • Built an autonomous AI agent using LangGraph that converts natural language requests into complete web applications (HTML, CSS, JS).
  • Designed a multi-agent workflow (Planner, Architect, Coder) with file-writing tools for end-to-end code generation.
  • Enforced structured LLM outputs using Pydantic for reliable disk-based file creation.

Education

Vellore Institute of Technology (VIT), Vellore

Integrated M.Tech Computer Science Engineering

CGPA: 9.13

2022 - 2027

Dayanand Junior Science College, Latur

High School – 12th

Percentage: 95.17%

2019 - 2021

Jawahar Vidyalaya, Jintur

Secondary School – 10th

Percentage: 93.40%

2019

Skills

Programming Languages

Python, SQL

Machine Learning & Data Science

Pandas, NumPy, Scikit-learn, LightGBM, XGBoost, Random Forest, TF–IDF, NLTK

Generative AI & LLMs

LangChain, LangGraph, RAG, Pydantic

Version Control

Git, GitHub

Familiar Technologies

C++, MLflow, CI/CD, Streamlit, Flask, Docker, DVC, DagsHub, AWS

Certifications

  • Introduction to Transformer-Based Natural Language Processing
    Credential ID: He2KFYSAyRj28w6Fgx1-aa
  • Building LLM Applications With Prompt Engineering
    Credential ID: v4r1blWQO-q2Ymc5eYfw
  • Oracle Cloud Infrastructure 2025 Certified Data Science Professional
    Credential ID: 1019469102SDSOCP
  • Oracle Cloud Infrastructure 2025 Certified Generative AI Professional
    Credential ID: 1019469102SDGAIOCP