- Developed an API to efficiently store simulation data into HDF files.
- Built data classes to improve monte carlo simulation pipeline.
- Performed unit testing and validation testing.
- Collaborated with the team to ensure code quality.
Hey there! I'm Satwik Kambham.
I'm on an exhilarating journey in the world of deep learning, passionately exploring the potential of AI models to redefine the boundaries of what's possible with data.
Projects
Papyrus
An elegant text editor based on atom written from scratch in tauri with Vite and Rust. Features include syntax highlighting, embedded terminal and LSP integration.
EN - HI Machine Translation
Machine Translation from English to Hindi on the OPUS100 dataset using the Transformer architecture. The model achieved good performance while having very fast inference times.
Underwater Scene Segmentation
Underwater Scene Segmentation with the Segformer architecture on the SUIM dataset. The project focuses on semantic segmentation of underwater imagery. The Segformer model demonstrates effective performance in delineating underwater scenes, making it a practical choice for underwater image analysis.
Low Light Object Detection
Low-Light Object Detection using DETR on the ExDARK dataset. The model, trained on 10 different low-light conditions, efficiently detects 12 object classes. The model was able to achieve state of the art results even in extreme low-light conditions by leveraging preprocessing techniques like CLAHE and Histogram Equalization.
Land Use Classifier
Image Classification on the UC Merced Land Use Dataset. The model was trained using various CNN architectures and the most efficient model (ResNet-18) was deployed on HuggingFace Spaces. The model achieved an accuracy of 98.5% on the test set.
Posts
Word Embeddings
Understanding the concept of word embeddings and the working of pre-trained embeddings like word2vec.
State Space Search - 8 Puzzle
Solving 8 puzzle using various state space search algorithms like A*, BFS, DFS, etc.
