Hey! I'm Rohan, an 22 year-old student with a passion for learning new things and food 🥑

Backend Engineer Intern on the Zero Trust Engineering Team at Shopify. Utilized technologies like Ruby on Rails, Typescript, React, GraphQL, Sideqik, and Cypress.

Achievements/Tasks Include:

• Added end-to-end tests to CI pipeline using Cypress and Typescript thereby automating the process of smoke testing dependency upgrades saving developer time.
• Created a process for automating involuntary employee exits by implementing a parallelized system of jobs using Ruby on Rails and Sidekiq which has helped over 100+ agents.
• Developed a service approval tracker for untrusted internal applications using React and Typescript increasing visibility into the service approval process for over 20,000 users.

Frontend Developer at Hack the North. Utilized technologies like Typescript, React, GraphQL, Context, and Tailwind.

Achievements/Tasks Include:

• Developed multiple web applications for Hack the North including hackthenorth.com, apply.hackthenorth.com, and my.hackthenorth.com amassing a cumulative 180k visitors.
• Used frameworks such as React, Context, Tailwind, and Apollo to create responsive and accessible web applications.
• Led the development of the Sponsor Dashboard project which enabled event sponsors to submit critical information and view attendee profiles.

Backend Engineer Intern on the Zero Trust Engineering Team at Shopify. Utilized technologies like Ruby on Rails, Typescript, React, GraphQL, Sideqik, and Cypress.

Achievements/Tasks Include:

• Developed a dashboard to display service proposal upvotes using Ruby on Rails, GraphQL, and React providing insight into voting metrics for over 3000+ services.
• Created a user access control dashboard using Ruby, GraphQL, and React that greatly reduces onboarding time for new agents.
• Implemented a session expiry notification feature for users submitting internal service proposals using Ruby on Rails and custom React hooks thereby reducing support time.

Software Developer Intern on the FX Marketplace Team at CIBC. Utilized technologies like MongoDB, Typescript, Node, Typemoq, Mocha, and Chai.

Achievements/Tasks Include:

• Completed a project to handle the reporting of all FINTRAC transactions to the Canadian FIU using TypeScript and Mocha.
• Wrote multiple APIs for the Simplii Gift Card (SGC) Microservice using TypeScript, Node, and MongoDB that is projected to service thousands of daily customers.
• Performed unit and integration testing for the SGC Microservice using Mocha, Typemoq, and Chai.
• Used OOP (Design Patterns), Agile development principles, and Git in the development of the FINTRAC and SGC Projects.
• Coordinated with multiple teams to discuss the FINTRAC app implementation, timeline, data processing, and testing.

H2O.ai is an AI based company that helps other businesses accelerate their adoption of cutting edge AI/ML into their products adding value to many facets of their business.

During my time at H2O, I learned a ton about data science and machine learning working under a senior data scientest - I got to learn about how various machine learning models function under the hood (Linear Regression, Decision Trees, Adaboost), how various machine learning metrics work, and completed an end to end machine learning project involving data manipulation, exploration, feature engineering, etc.

Achievements/Tasks Include:

• Learned a ton about how various ML models function (Linear Regression, Decision Trees, Adaboost).
• Worked on a time series ML task using H2O-3.
• Performed benchmarking and user analysis on the Driverless AI Platform.

You can check out the colab notebook here.

MarketLabs is a digital marketing company that provides website creation, social media management, and SEO services to small businesses.

As the founder of MarketLabs I was tasked with getting funding, finding clients, and marketing the business. As for funding, I applied and got accepted to the Town of Richmond Hill Summer Company Program which provides high school and college students with appropriate new business funding ($3000) and mentorship. Once I had gotten the money I started to market the business in hopes of finding clients, however, this proved tougher than I anticipated. While it was a painstaking process that involved messaging small business owners, going to business after business, and networking and pitching my business at trade shows, I was able to find 2 clients that resulted in a total of $2000 in profit for the summer of 2019.

Achievements/Tasks Include:

• Applied & Received $3000 in funding from the City of Richmond Hill as a part of the Summer Company Program.
• Consulted with local businesses about online marketing.
• Proposed marketing strategies & created websites for clients.
• Produced $2500 in profit over the summer of 2019.

The company website is here.

This project is an account portfolio viewer for Questrade clients. This project allows users to understand the contents of their portfolios, use interactive graphs, get real-time stock + options pricing data, and manage account orders easily.

You can find the link to the live Questrade Portfolio Viewer application here and the repo here.

How it was built:
This project was built using React + Redux on the frontend and Node + MongoDB + Express on the backend.

• Used the OAuth2 Standard for user authentication. Implemented custom Express middleware to preemptively fetch new access tokens upon expiry.
• For security purposes AES Encryption, Passport.js, and Cookie-Session were used to maintain a user's session securely.
• Used MongoDB along with Mongoose to store user data.
• Utilized a client side proxy along with Axios to forward appropriate requests between the frontend and backend.
• Redux + Redux Thunk were used when writing action creators and reducers to fetch data from the backend and update the client side Redux Store.
• Utilized the Nivo Rocks graphing library to create interactive graphs.
• Utilized a mix of both functional and class based components along with component hooks and lifecycle methods.
• Utilized Web Sockets on the client side to fetch real-time pricing and orders data.
• Used Semantic-UI along with custom CSS and CSS Media Queries to style the site and create responsive behaviour.

This is a C++ implementation of the Straights card game.

You can find the link to the repo here.

Overview:

• This project follows an Object-Oriented design structure, using classes and objects to organize information and maintain an extensible codebase for future upgrades. This design follows the Model-View-Controller design method.
• The project consists of multiple classes, the main ones being StraightsModel, Player, and View. The StraightsModel class is the main hub of functionality for the game, and handles the information regarding players, the game pile, the deck, and is where the main portion of the game is run. The Controller and View classes both interact with this class and serve as the functionality that allows the StraightsModel class to interact with players. The Controller is an abstract base class, with its children implementing specific methods pertaining to how the game should receive input from the user. The View class is also an abstract base class that interacts with the StrightsModel class and updates information to the implemented method desired. Moreover, the View class is an Observer and the StraightsModel class is the subject in an Observer pattern intended to make it easy to add more views to the application in the future, making it more resilient to future change.

Design:

• This project uses a myriad of Design patterns that help with solving certain design problems and make the code easier to modify in the future. The first main design pattern that this project implement is the Model-View-Controller (MVC) design pattern. Moreover, to implement the ability to add more views, I have implemented the Observer design pattern to keep track of the different views the application has, and the Model class can update them accordingly. Moreover, the Strategy design pattern was also used in the Controller classes in order to make receiving user input from any device is easy.
• The use of multiple classes and Design patterns make this project quite resilient to change and easy to upgrade and extend in the future as the needs of potential players change. A developer can easily add another view to this program by extending from the View class and adding it using the addView method in the Model class.

Coronaanalytics is a interactive dashboard that contains up-to-date COVID-19 data, plots, and news making it easier to visualize COVID-19 in different countries, as well as the global impact. I developed this project using Python, specfically Dash and Plotly for the backend and interactive portions of the site, and HTML, CSS, and JS for the general design of the site.

You can find the link to the repo here.

RapBot is a machine learning application that generates rap lyrics using an LSTM network on a character based prediction model.

The training of the machine learning model was done using Keras and Python with a datset of many Eminem lyrics, then the weights were stored in a .h5 file. The model works by predicts the next character given string of length 300 characters. The webserver portion of the site was built using Flask.

You can find the link to the repo here.

This project was a part of the Kaggle COVID-19 Global Forecasting competition. In this competition, you are asked to predict future cases and fatalities of the virus for all countries around the globe. This project was especially challenging given the lack of information given, however, through the adding of external data such as population density, Malaria cases, and Google's COVID-19 Community Mobility Report data I was able to achieve a decent score in the competition.

You can find the link to the colab notebook here.

Snapfood is an IOS app that uses Machine Learning to classify different foods. The app then returns nutritional Information, recipes, and descriptions of the classified food. The app was developed using Swift and Core ML. And yes, this app was inspired from the TV show Silicon Valley 😅

You can find the repo here.

ImageRepo is a Image Repository service, where users can buy and sell images easily.

Users can first login/sign up to the application using their Facebook accounts, then they will be sent to a page where they can browse images for sale, add credits to their account, upload their own images, and more. Myriad technologies were used in the development of this project.

The front-end was built using JavaScript, React, Redux, and Bootstrap-React for styling. The backend was built with Node, Express and MongoDB.

Along with the main technologies used, I also made use of concepts such as OAuth to implement the "Sign in with Facebook" feature, a pre-trained ML model that classifies images to implement an image search based on characteristics, and an advanced image rendering feature that mirrors how images are rendered in Google images.

You can find the link to the live ImageRepo application here and the code here.

Flamingo is a networking application that makes it easy to get someone's LinkedIn profile through bluetooth. Users who would want to exchange LinkedIn profile information at a conference or event would simply hold their phones together closely with Flamingo open, and their LinkedIn profiles would be transferred to each other's phones.

Using the LinkedIn API and Swift, my hackathon team and I built this application. This project was a finalist at MasseyHacks.

Software Engineer Intern at Kikoff (Jan 2025 - Apr 2025, San Francisco).

Achievements/Tasks Include:

• Built an account reactivation system using Rails & Flutter, driving a 58% boost in returning subscribers & cutting support requests 22%
• Built a system for OFAC screening with OpenSanctions using Rails & Sidekiq, including scoring user PII to sanctions list, generating alerts, and case-management workflows, saving ~$750K/year
• Optimized network call latency in Rails by moving non-essential network calls out of database locks/transactions via after-commit callbacks, reducing lock durations and minimizing rollback failures
• Created a Rails & Flutter survey framework with nested questions, randomized options, multi/single-select, and A/B testing, enabling configurable in-app surveys and analytics

Software Engineer Intern at Newfront (YC W18) in San Francisco.

Achievements/Tasks Include:

• Optimized app performance by implementing virtualization and debounced search for multiple critical components using Typescript and React, resulting in a ~40% reduction in load time
• Engineered an underwriter quote cloning system with NestJS and AWS SNS/SQS, implementing event-driven async file replication, streamlining workflows for over 400+ producers
• Implemented distributed error tracing for the event outbox system using Typescript and Datadog, improving system observability and debugging efficiency

Software Engineer Intern at Empire.

Achievements/Tasks Include:

• Created a deployment workflow utilizing GitHub Actions and GCP to automate production and feature branch deployments, achieving a ~40% reduction in deployment times
• Engineered a voting system for technology proposals using Python and Firebase, leveraging hashed IDs and an optimized database schema with denormalization to minimize database queries
• Implemented an authentication and RBAC system using Auth0 and React with custom post-login actions to sync user Google group data, thereby enhancing user access control

Flow is a personalized AI assistant that ingests user data from multiple platforms, that enables an LLM powered chat based on a user's personal information:

• Implemented a sophisticated RAG system including intelligent retrieval with hybrid vector search, Cohere reranking, and an AI-driven chunk organizer that groups semantically related content from different sources while considering temporal relevance
• Engineered a vector indexing pipeline using Qdrant, with sparse and dense vectors, leveraging FastEmbed for embeddings and LlamaIndex for semantic chunking
• Designed a context window management system that prevents overflow by tracking model-specific token limits, and allocating budget between the system prompt, the conversation history, and the retrieved chunks
• Added a distributed task queue with Celery, enabling parallel processing of document syncing with rate limiting, fault tolerance, and automated retry mechanisms
• Built a full stack app with Python, GraphQL, and FastAPI on the backend and React, Vite, and Tailwind on the frontend
• Links:
  • Live production app: https://tryflow.dev/
  • Demo video: https://www.loom.com/share/9f27c67a72224c07903fc9341071be5d?sid=e9eadaf4-0607-48ec-9d61-71612ee46d96

• Created a text processing pipeline using DistilBERT to generate embeddings for pros, cons, and headlines separately, preserving the semantic context of each component
• Engineered features from review text including length metrics, sentiment scores using TextBlob, and firm-specific statistics to enhance prediction accuracy
• Designed a neural network with parallel processing pathways that handle different input types (text embeddings and categorical features) before combining them
• Implemented training optimizations including early stopping, learning rate scheduling, and validation monitoring to improve model performance
• Achieved 2nd place in a class competition
• Links:
  • Code: https://github.com/rohanrav/glassdoor-rating-predictor

This research project examines bias in large language models across racial, ethnic, and gender dimensions.

Research Components:

• Tested multiple LLMs (GPT-3.5, Llama2, Gemini Pro) to uncover potential racial and gender biases by analyzing responses to identical prompts across different demographic profiles
• Applied NLP techniques to analyze sentiment, word frequency, and response patterns in LLM outputs, creating a data-driven approach to identify subtle biases in AI-generated content
• Built statistical analysis workflows using ANOVA and cosine similarity metrics to quantify variations in how LLMs respond to different demographic identifiers
• Created visualizations that clearly showed bias patterns (or lack thereof) in LLM responses, helping communicate complex findings in an accessible format
• Compared censored vs. uncensored LLM versions to evaluate the effectiveness of current bias mitigation techniques in commercial AI systems
• Links:
  • Report website: https://cs492-llm-bias-report.vercel.app/