Skip to content
View Vansh-A1's full-sized avatar
3 followers · 3 following

Block or report Vansh-A1

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
Vansh-A1/README.md

Hi, I'm Vansh Joshi

AI student · Building from first principles · Learning through experiments

I study artificial intelligence by implementing its building blocks and exploring how model architectures behave in practice. My work spans language models, generative modeling, computer vision, and agentic AI.

Currently building: LLM From Scratch — progressing from tokenization and embeddings toward causal attention, Transformer blocks, and a complete training pipeline.

Start here

Project What I'm exploring
LLM From Scratch Language-model foundations: tokenization, context/target pairs, and embeddings. Attention and training are the next stages.
Computer Vision from First Principles Classical vision through Harris corners, Fourier-based hybrid images, and image moments.
Diffusion Models An unconditional DDPM implemented in PyTorch and a class-conditioned MNIST workflow using Diffusers.
Thermal Super-Resolution Experiments with 8× thermal image reconstruction, residual networks, edge information, and frequency losses.
Underwater Vision Image enhancement connected to tracking and segmentation in underwater video.
SpendGuard A prototype for evaluating agent spending requests with policy, intent, evidence, and risk checks.

How I work

  • Understand the mechanism. Connect an equation or architectural choice to a readable implementation.
  • Build incrementally. Turn small components into larger systems, with explicit milestones.
  • Learn through experiments. Inspect tensor shapes, model outputs, and failure cases.
  • Document progress. Explain what is implemented today and the next technical questions to investigate.

Tools I use

Python · PyTorch · NumPy · OpenCV · Matplotlib
Also exploring Hugging Face tooling, language-model workflows, and agentic AI.

My repositories include ongoing learning projects and applied experiments. I welcome technical feedback, thoughtful questions, and focused contributions through their GitHub issues.

Pinned Loading

  1. LLM_FROM_SCRATCH LLM_FROM_SCRATCH Public

    A language model built from scratch in PyTorch, with readable Transformer components, streaming training, and separate learning exercises.

    Python

  2. computer-Vision-from-first-principle- computer-Vision-from-first-principle- Public

    Classical computer vision explained through Python, NumPy, and OpenCV: Harris corners, frequency-domain hybrid images, and orientation from image moments.

    Python 4

  3. SpendGaurd SpendGaurd Public

    An agent spending-control prototype combining policy checks, intent alignment, evidence verification, behavioral risk, and human review.

    Python 4

  4. Under-water-image-enhancement-and-tracking- Under-water-image-enhancement-and-tracking- Public

    Underwater video enhancement connected to single-object tracking and segmentation, with annotated outputs and evaluation utilities.

    Python 4

  5. Thermal-SuperResolution-8x Thermal-SuperResolution-8x Public

    8× thermal image super-resolution experiments with CNNs, residual networks, GANs, edge-aware reconstruction, and frequency-domain losses.

    Python 3