Skip to content
View Ultron09's full-sized avatar
🗿
Creating
🗿
Creating
14 followers · 30 following

Achievements

Achievement: Pair Extraordinairex2Achievement: Pull Sharkx3Achievement: YOLOAchievement: Quickdraw

Achievements

Achievement: Pair Extraordinairex2Achievement: Pull Sharkx3Achievement: YOLOAchievement: Quickdraw

Highlights

Organizations

@AirBorneAI

Block or report Ultron09

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Ultron09/README.md

Suryaansh Prithvijit Singh

Founder @ AirBorne · AI Researcher & Systems Software Engineer
Bangalore, India · LinkedIn · X (@Suryaansh07) · connect.singha@gmail.com

I build and research deep learning systems from mathematical first principles—specializing in continual learning architectures, custom neural models built from scratch, and high-performance backend systems.

🔬 Selected Research & Architectures

Neural & Systems Architecture Schematics

  • ANTARA // NeurIPS Testbed — Evaluation testbed for continual learning under Split CIFAR-100. Implements sequential learning across 10×10 tasks without raw data rehearsal, benchmarking OGD and RGW dual-system architectures against EWC, ER, and A-GEM.
  • Numpy-Transformer — Pure NumPy implementation of an autoregressive GPT built from scratch. Features manual backpropagation, multi-head self-attention, and custom Adam optimizer with zero high-level framework dependencies.
  • Vision-Transformer-Diffusion — Minimal ViT + DDPM diffusion pipeline built from scratch in NumPy, featuring a simplified ViT-UNet and AdaLN time conditioning.
  • MSOPT & LOB — Multi-Scale Overlapping Pattern Tokenization and Limit Order Book (LOB) quantitative modeling for financial time-series forecasting.
  • C_Language_model — Minimal, dependency-free Transformer architecture implemented in pure C.

🛠️ Core Stack

  • Systems & Languages: C++, C, Python, TypeScript, CUDA, Linux, Bash
  • Machine Learning: PyTorch, NumPy, JAX / Flax, OpenCV, Scikit-Learn
  • Infrastructure & Backend: Docker, Redis, Next.js, Node.js, PostgreSQL, Supabase

Tech Stack Marquee

Pinned Loading

  1. BrainTumor BrainTumor Public

    Detecting and Classifying Brain Tumor by Creating a CNN model based on VGG16

    Jupyter Notebook 1

  2. Numpy-Cnn Numpy-Cnn Public

    Advanced CNN Training from Scratch with NumPy. A modular, educational Deep Learning framework implementing Convolutional Neural Networks without high-level libraries. Features 15+ layer types, 8 ac…

    Python 2 1

  3. Numpy-Transformer Numpy-Transformer Public

    A pure NumPy implementation of GPT built from scratch for educational purposes. Features manual backpropagation, multi-head attention, and a custom Adam optimizer with zero deep learning frameworks.

    Python 2

  4. LeetCode LeetCode Public

    Optimized C++ solutions and algorithmic pattern documentation for LeetCode data structures and competitive programming.

    C++ 1

  5. huggingface/transformers huggingface/transformers Public

    🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.

    Python 165k 34.5k

  6. sqliteai/warp sqliteai/warp Public

    Run the full 2.78-trillion-parameter Kimi K3 model or GLM-5.3-Flash beyond available RAM by streaming activated weights directly from NVMe. A dependency-free, embeddable C inference engine.

    C 2.4k 178