Kashish Verma | Portfolio

About Me

Who I Am

I’m a computer science student and full-stack developer with a growing interest in AI. I work with C++, and build modern web apps using Node.js, React, and MongoDB. I enjoy creating scalable, user-friendly solutions that make an impact.

Recently, I fine-tuned a BERT model for sentiment analysis and published it on Hugging Face 🤗. I also built WhiskersHaven, an AI-powered cat adoption platform using OpenAI GPT-3.5, MongoDB Atlas Vector Search, role-based access, Stripe payments, and real-time analytics.

I'm currently exploring Transformer architectures, LangGraphs, and how to make machines “think” (or at least pretend to). Outside of code, I love writing tech blogs, reading romantic fiction, cooking, and discovering new food spots.

Download Resume

Work Experience

My professional journey and the companies I've worked with

Jan 2025 - May 2025

Software Developer

Architech • Toronto, ON

Developed AI-powered customer support automation using LangGraph/ReAct for diverse query handling and agentic workflows.

Key Achievements:

Improved LLM performance via Semantic Search with OpenAI embeddings/pgvector/HNSW
Implemented human escalation and automated CI/CD pipelines for Python/React apps
Utilized Azure DevOps and GitHub Actions for continuous integration and deployment

Python

React

LangGraph

OpenAI

Azure DevOps

GitHub Actions

Jan 2024 - Present

Math Instructor

Mathnasium • Toronto, ON

Provide personalized math instruction to students of various ages and skill levels, helping them improve their understanding and performance in mathematics.

Key Achievements:

Developed customized learning plans for students based on their individual needs
Implemented engaging teaching methods to make math concepts more accessible
Maintained regular communication with parents about student progress

Teaching

Mathematics

Communication

Problem Solving

Mentoring

My Projects

Here are some of the projects I've worked on

Emotion-Classify-BERT

BERT

Transformers

PyTorch

Hugging Face

NLP

Machine Learning

Emotion Classification

Fine-tuning

A fine-tuned BERT model for emotion classification achieving 94.01% accuracy. Built on bert-base-uncased and trained on the mteb/emotion dataset to classify English text into six emotion categories: sadness, joy, love, anger, fear, and surprise. Features fast inference, easy integration with Hugging Face transformers pipeline, and robust performance on emotion detection tasks.

Code Try Model

WhiskersHaven

Node.js

MongoDB Atlas

OpenAI GPT-3.5

Vector Search

JWT

OAuth

Stripe

Cloudinary

Nodemailer

Tailwind CSS

Chart.js

Google Maps API

A multi-modal AI adoption assistant leveraging Retrieval-Augmented Generation (RAG) by combining OpenAI GPT-3.5 and MongoDB Atlas Vector Search for semantic pet matching. Features secure JWT/OAuth authentication, shelter and user management, real-time analytics dashboards, Reddit-style nested comments, and media handling via Cloudinary. Fully integrated with Stripe for payments and Nodemailer for email notifications. Built with a responsive Tailwind CSS frontend delivering an intuitive user experience.

Code Live Demo

Lego Collection App

Node.js

Express

PostgreSQL

Sequelize ORM

RESTful API

Tailwind CSS

A feature-rich application for Lego enthusiasts to manage and explore their collections. Implements a PostgreSQL database with Sequelize ORM for efficient data management, RESTful API endpoints, and a responsive UI with Tailwind CSS. Includes advanced filtering, sorting, and search capabilities.

Code Live Demo

Library Management System

C++

OOP

Data Structures

File I/O

Console Application

Binary Search Trees

A robust C++ application for managing library resources, implementing advanced object-oriented programming principles. Features include comprehensive book tracking with search functionality, user management with borrowing history, and an efficient data structure design for optimal performance.

Code View Project