Recently read TikTok's whitepaper on Monolith, their real-time recommendation system and was so fascinated. So I rebuilt parts of it on a small scale. This is a pipeline built using Kafka for streaming and PySpark for transformation.

RAG-based Ad Generation Platform

Architected and built a complete ad generation platform for RMIT University's Marketing Team. The platform allows users to analyse competitor ads and generate new ads using GPT4o and RAG. Single handedly built the auth system, competitor analytics' frontend and backend followed by full RBAC implementation. Deployed everything on AWS using services like RDS, Lambda and ECR for containers.

Built an end-to-end MLOps project for a time series forecasting model using AWS. The pipeline includes data ingestion, preprocessing, model training, and deployment. The model is trained on a dataset of 20000+ samples and achieves an RMSEs of 25-30 on average.

Built a NLP/Logistic Regression model to classify job postings into categories. The model was trained on a dataset of 700 job postings. The deployed model achieved an accuracy of 91%. Also built a nice React frontend around it.

Built a cloud storage application using NextJS and .NET with AWS. Users can sign up, login and start uploading files of any type. Authentication runs on AWS DynamoDB and File Uploads run on S3. Upon user creation, a folder is created in an S3 bucket. This folder is specific to the user and any file they upload will reside in this folder. Implemented JWT auth as well.

Created a data pipeline for streaming state-wise (Australia) electricity generation and pricing data from AEMO to a PostgreSQL database using Kafka.