Projects
FocusStream
Distributed AI Video Processing Pipeline · Java · Python · Kafka · Kubernetes · MinIO
FocusStream is a distributed AI video processing pipeline built on a twin-cluster Kubernetes topology with separate CPU and GPU control planes to limit compute cost for active workloads. A presigned-URL task payload strategy routes video directly between MinIO object storage and compute workers, and KEDA autoscaling on Kafka consumer lag scales AI worker pods from zero on demand, eliminating idle GPU infrastructure cost. Built with Java, Python, Apache Kafka, Kubernetes, MinIO, KEDA and Docker.
Deployable SeamlessM4Tv2 API Endpoints
Python · Docker
Docker-containerized, deployable API endpoints for Meta's SeamlessM4Tv2 model that generate text-to-speech and speech-to-speech translation audio. Built during the MCREU research internship at Penn State to power real-time multilingual communication for the Communicator.AI iOS app, deployed with Gunicorn and Flask on AWS EC2 with an S3 bucket for audio storage. Built with Python, Docker, Flask, Gunicorn, AWS EC2, AWS S3 and NLP.
Deployable SeamlessM4Tv2 API Endpoints on GitHubVision Transformer Implementation
Python · PyTorch
A Vision Transformer (ViT) image classification model trained on standard benchmark datasets, implementing the patch-embedding and multi-head self-attention architecture from scratch in PyTorch with configurable depth and head count. Includes training scripts, evaluation metrics and visualization of attention maps. Built with Python, PyTorch, Transformers and computer vision techniques.
Vision Transformer Implementation on GitHubCommunicator.AI iOS Application
Python · Swift · AWS
Communicator.AI is an iOS application enabling seamless real-time translation and communication between languages, built as part of the Penn State MCREU Scholars research program. It integrates Meta's SeamlessM4T-v2 NLP model via a Flask REST API containerized with Docker and deployed on AWS SageMaker, with TestFlight used for beta testing and feedback from the Hershey Medical collaboration. Built with Swift, Python, AWS, Docker, Flask, NLP and iOS.
Communicator.AI iOS Application on GitHubmodel.aio
Python · LiteLLM · FastAPI · Next.js · Under Development
model.aio is a chatbot interface enabling simultaneous multi-model inference — querying GPT-4, Claude, Gemini and over 1000 supported models side by side from a single unified UI. Under active development, it abstracts provider SDKs behind a common interface, streams responses in parallel and lets you compare outputs at a glance. Built with Python, LiteLLM, React, token streaming and FastAPI.
Stock Graph Visualizer & Data Aggregator
Java · Spring Boot · Electron
A Spring Boot desktop application for visualizing stock trends from historical market data. The Spring Boot backend fetches and processes OHLCV data from a financial API while an Electron frontend renders interactive line charts, supporting multiple tickers and configurable date ranges. Built with Java, Spring Boot, Electron, JavaScript and a REST API.
ComfyUI Video Generation Module
Python · Next.js · Runpod API · Docker
An end-to-end application integrating the Runpod API to automate video generation workflows within ComfyUI. A Next.js frontend pairs with a Python worker that includes a custom ComfyUI workflow parser for determining inputs dynamically, reducing the manual steps required to run open-source video generation models on RunPod VMs. Built with ComfyUI, Next.js, Docker, the Runpod API and FastAPI.
ComfyUI Video Generation Module on GitHubResume Builder - powered by n8n
Python · n8n · Docker · Under Development
A barebones n8n workflow that helps make a resume ATS-parseable — an experiment with no-code tooling. Given details about a job posting, it produces a tailored resume for that specific role using a simple OpenAI API call, saving the manual back-and-forth of switching tabs and writing prompts. Built with n8n, Next.js, Docker and the GPT API.
Resume Builder - powered by n8n on GitHubLabel Automation System
Python · PyQt6 · FastAPI · ShipeurX & Affiliated
An end-to-end order fulfillment desktop application built with PyQt6 and FastAPI, integrating DHL shipping APIs for automated label generation and real-time Ecwid storefront order status synchronization. Reduced order processing time by 83% — cutting fulfillment from 6 hours to 1 hour — by eliminating manual label reconciliation across thousands of SKUs. Built with Python, PyQt6, FastAPI, the DHL API and the Ecwid API.
Number Plate Recognition Pipeline
Python · OpenCV · YOLO model family · NTPC Chattisgarh, India
A production-deployed license plate detection pipeline using YOLO and OpenCV for automated truck checkpoint verification at NTPC Chattisgarh, India. Curated and manually annotated a custom training dataset of truck license plates, achieving near-perfect detection accuracy under standard lighting conditions. Built with Python, OpenCV, the YOLO model family, model finetuning and a REST API.
