Projects

Things I've built, from diffusion models to IoT hardware. Every project has its source on GitHub.

Scroll down to flip through the projects

Zone-Music

A native Android music player written in Kotlin. The interface is built entirely in Jetpack Compose and audio playback runs on ExoPlayer, the same media engine many Android music and video apps use. It is my most recent project.

  • Kotlin
  • Jetpack Compose
  • ExoPlayer
  • Android
  • Gradle (Kotlin DSL)
  • Declarative UI: screens are composed with Jetpack Compose instead of XML layouts
  • Media engine: ExoPlayer handles audio decoding and playback
  • Modern Android stack: 100% Kotlin, built with Gradle Kotlin DSL scripts
View source on GitHub

Stego Suite

A multi-format steganography toolkit for hiding data in images, audio and video using LSB encoding. It does not include encryption.

  • C++
  • Python
  • LSB encoding
  • FFmpeg
  • Image steganography: hides any file inside a PNG using the RGB channels, with a custom header for reliable extraction
  • Audio steganography: Mid/Side channel LSB encoding embeds secret audio in cover audio, with lossless 16-bit PCM WAV output
View source on GitHub

Exam-Compass

A retrieval-augmented generation (RAG) study tutor for NEET and JEE preparation that indexes NCERT textbooks.

  • Python
  • ChromaDB
  • sentence-transformers
  • LLM
  • RAG
  • Grounded answers: study material is embedded with sentence-transformers and retrieved from ChromaDB to ground the LLM
  • Low-cost inference: the LLM backend runs on Google Colab behind a Cloudflare Tunnel, with no dedicated server
View source on GitHub

PDF Watermarking Tool

A Python GUI for automated PDF branding and merging. It powers the document workflow at KTU Special.

  • Python
  • Tkinter GUI
  • PyPDF2
  • Automation
  • Clickable hyperlink watermarks: embed interactive URLs
  • Batch processing: handle 100+ files at once
  • Image watermarks: scale and transparency control from 0 to 100%
  • Front and appended pages: add custom cover and end pages
View source on GitHub

Zero-Shot Low-Light Image Enhancement (published research)

A CLIP-guided semantic diffusion pipeline that brightens low-light images with no paired training data and no fine-tuning, with a Gradio demo hosted on Colab.

  • PyTorch
  • Stable Diffusion v1.5
  • CLIP
  • Gradio
  • Semantic guidance: CLIP ViT-B/32 steers Stable Diffusion v1.5
  • Consistency filter: rejects outputs that drift from the scene (CLIP cosine similarity above 0.85)
  • Faster sampling: DDIM and FP16 cut sampling from 1000 steps to 40-50
  • Evaluation: LOL and ExDark with PSNR, SSIM, LPIPS and MUSIQ: 15.56 dB and 0.729 SSIM vs 15.40 dB and 0.515 for a retinex baseline, on the evaluated samples
View source on GitHub

IoT Crop Recommendation System

A smart agriculture system where ESP8266/ESP32 sensor nodes capture real-time soil and environmental data and a deep neural network recommends crops to farmers.

  • ESP8266 / ESP32
  • Deep neural network
  • Python
  • TensorFlow
  • Real-time monitoring: soil moisture, pH and temperature sensors
  • DNN prediction model: 95%+ accuracy in crop suitability
  • Farmer dashboard: clear visualisation of recommendations
View source on GitHub