Skip to content
View rakeshvar's full-sized avatar
100 followers · 48 following
  • Stanford University
  • London

Achievements

Achievement: Arctic Code Vault ContributorAchievement: Starstruckx2Achievement: Pull Shark

Achievements

Achievement: Arctic Code Vault ContributorAchievement: Starstruckx2Achievement: Pull Shark

Block or report rakeshvar

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.

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
rakeshvar/README.md

“In the beginning there was neither existence nor non-existence.” — Ṛgveda

👋 Hello

You can Call me Cash. Stanford gave me a PhD in AI.

  • For 13 years I have been working on Deep Learning
  • Language modeling has always been a big draw for me, be it Optical Character Recognition or text prediction for Sanskrit.

🖼️ Diffusion Modeling

Current focus is on more advanced stuff, specialising in Transformers based Diffusion Models.

  • Permutation-invariant diffusion on sets with fractal geometry, a hybrid of GPT, Diffusion and World Models
  • An independent research building permutation-invariant DDPM/DDIM models for aperiodic chaotic tilings on fractal manifolds
  • Penrose Diffusion

🎒 LLMs

Also training nano-GPT style models on Sanskrit text.

  • Transformer & LLM architectures from scratch
  • Repo Sanskrit GPT

👨‍🎓 PhD Work

Implements first-order methods from my PhD thesis on L₁-regularized infinite-dimensional convex neural networks (10★)

  • AnyBoost.jl – Julia package for boosting any loss, activation, and constraint.

🏰 Foundational Work

Extensively on Optical Character Recognition

  • Universal script-agnostic OCR (CNN + LSTM + CTC)
  • Just with a basic set of fonts and sample corpus data, we can build an complete segmentation free OCR system

📍 Selected Open-Source Projects

  • Chamantt OCR – Universal OCR for any language/script (Hindi, Arabic, Telugu, etc.)
  • Banti OCR – End-to-end Telugu OCR framework. CNN-based, font-free system with n-gram language modelling, glyph stitching & erasure recovery. Extensible to other Indic scripts. 53★ (arXiv:1509.05962)
  • Rnn CTC – One of the earliest GitHub RNN+CTC sequence learning libraries (70+ stars)
  • Theanet – Keras-style CNN library in Theano with full augmentations & regularizations. Built this before Keras was a thing!

🏛️ Background

  • PhD Statistics, Stanford (advisor: Prof. Trevor Hastie)
  • Thesis: Path algorithms for ℓ₁-regularized infinite-dimensional convex neural networks
  • 1750+ hours classical yoga teacher training

🛠️ Skills

Languages & Frameworks
Python PyTorch TensorFlow JAX/XLA Julia C++ CUDA

Languages
Spanish • Sanskit • Hindi • English • Telugu • Kannada • Tamil

Currently open to collaboration or roles at the frontier of generative AI research.
Reach me at my github id@gmail.com

Pinned Loading

  1. penrose_diffusion penrose_diffusion Public

    Permutation Invariant DDPM/DDIM Model on Sets} to sample-without-repetition chaotic aperiodic tilings on zero-measure fractal manifolds. Built a plug-and-play architecture for Transformer Denoisers…

    Python 1

  2. SanskritGPT SanskritGPT Public

    Training a Generatively Pre-trained Transformer on Sanskrit data.

    Python

  3. rnn_ctc rnn_ctc Public

    Recurrent Neural Network and Long Short Term Memory (LSTM) with Connectionist Temporal Classification implemented in Theano. Includes a Toy training example.

    Python 221 81

  4. AnyBoost.jl AnyBoost.jl Public

    A julia based machine learning package for boosting any loss, activation and constraint.

    Julia 10