Projects

OrbitAI

PythonC#Three.jsPyTorchFlaskONNXAWS EC2Unity
Check it out
OrbitAI preview

An AI-powered satellite traffic management system using hybrid neural networks to predict and prevent collisions in increasingly crowded low Earth orbit.

  • Implements a Gated Recurrent Network (GRU) to predict real-time satellite position trajectories over 180-step sequences using synthetic LEO satellite data.
  • Utilizes Graph Neural Networks (GNN) to coordinate satellites as dynamic nodes and predict high-risk interactions between spacecraft in orbital space.
  • Features a Unity-based 3D orbital simulator that visualizes collision scenarios and avoidance maneuvers with live AI predictions and trajectory modeling.
  • Deploys the machine learning model on AWS EC2 with Flask server and WebSocket communication for real-time inference between simulation and AI backend.
  • Addresses the critical problem of space debris collision avoidance as satellite populations are projected to exceed 100,000 active satellites in LEO by 2030.
  • My favorite project so far :-)

Mosaic -- In progress...

PythonPyTorchNode.jsPostgreSQLDockerKubernetesD3.js
Check it out

A distributed multilingual NLP pipeline for real-time global sentiment tracking.

  • Built a distributed multilingual NLP pipeline that ingested and processed 50k+ daily news headlines across English, Spanish, and Chinese via Kafka streaming, enabling real-time global sentiment tracking.
  • Applied transformer-based sentiment models, FinBERT + XLM-RoBERTa, fine-tuned on financial text, achieving a 20% higher correlation with short term equity moves.
  • Developed lag-correlation analytics and interactive dashboards in D3.js to visualize sentiment-market dynamics, providing insights into news-driven shifts within 15 minutes of release.

Certis

GolangRustPostgreSQLReactDocker
Check it out

A tamper-proof blockchain system built in Go for recording and verifying event attendance certifications with cryptographic integrity and web interface.

  • Implements a complete blockchain with SHA-256 hashing, block validation, and cryptographic chain linking to ensure data immutability.
  • Features a web-based submission system with HTML forms for attendees to register their event participation through a Go HTTP server.
  • Includes persistent JSON storage with automatic blockchain state saving and loading between application restarts.
  • Provides QR code generation functionality to create quick links for easy access to the certification submission forms.
  • Designed with extensible architecture supporting planned features like PostgreSQL integration, Rust validation engine, and React frontend

MOODSIC

PythonDjangoTypescriptReactOpenCV
Check it out
MOODSIC preview

Detects a user's emotional state through facial expression analysis using a webcam and recommends a song that matches or elevates their mood by integrating with Spotify's API.

  • Frontend developed using React for a responsive UI with live webcam integration.
  • Backend built with Flask (Python) to handle routing, process emotion data, and interact with Spotify’s API.
  • Emotion detection implemented using OpenCV and a deep learning-based facial emotion recognition model.
  • Spotify Web API used for authenticating, accessing playlists, and retrieving track details based on mood and genre.
  • Challenges included emotion detection accuracy, managing Spotify API limitations, ensuring real-time performance, and achieving cross-browser compatibility.

Gitissues

Rust
Check it out

A command-line tool for listing, creating, and managing GitHub issues directly from the terminal.

  • Fetches and displays a list of issues from a GitHub repository.
  • Supports filtering issues by status, labels, or assignee.
  • Allows creating new issues without visiting GitHub web interface.
  • Provides summary statistics like open vs closed issues and issue count per label.
  • Designed as a lightweight CLI tool to streamline issue management for developers.

ImgCap

PythonPyTorch
Check it out

A PyTorch-based image captioning model using CNN encoder and LSTM decoder architecture trained on the Flickr8K dataset to generate natural language descriptions of images.

  • Implements an EncoderCNN class using pre-trained Inception v3 to extract visual features from input images and map them to embedding space.
  • Features a DecoderRNN class with LSTM layers that processes image features and generates sequential word predictions for caption generation.
  • Combines encoder and decoder in a unified pipeline (crpipe) class that handles end-to-end training and inference for image-to-text translation.
  • Includes training infrastructure with configurable hyperparameters, checkpoint saving/loading, and TensorBoard integration for monitoring training progress.
  • Provides caption generation functionality with beam search capabilities and vocabulary mapping to convert token predictions back to readable text.