Skip to content
View ItzAditya43's full-sized avatar
🎯
Peak Performance
🎯
Peak Performance

Block or report ItzAditya43

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
ItzAditya43/README.md

Aditya Pandey

Typing banner

I build local-first AI tools and Linux desktop apps. Most of it runs on your own machine, with no cloud and no API keys.

🏯 Walk through my 3D portfolio · a floating Japanese village in Three.js with day/night, seasons and shrines

X / Twitter   LinkedIn   Kaggle   LeetCode   CodeChef   Email


🔩 From scratch, not from a tutorial

  • 🧠 A language model: tokenizer, transformer, training loop and KV-cache inference, trained on a single 4GB GPU
  • 🛰️ A network bypass layer: per-app WireGuard, encrypted DNS and TLS-record fragmentation, in Rust
  • 🤖 A coding agent: runs entirely on local models, nothing leaves the machine
  • 🎛️ A Linux control plane: processes, cgroups, systemd, Docker and GPU behind one app
  • 🔮 A world model: learns physics with no reward signal and plans by imagining futures

🏆 Featured Projects

Project What it is Stack
Zenith · live Self-hosted AI workspace on your own Ollama models: chat, a coding agent, deep research, long-term memory, image gen and voice. Nothing leaves your machine. Python · FastAPI · React · Tauri
Zenith-AI A small language model built from scratch: BPE tokenizer, RoPE/GQA transformer, training loop and KV-cache inference, trained end to end on a single 4GB GPU. Python · PyTorch
Detour Per-app WireGuard tunneling, system-wide encrypted DNS and TLS-record fragmentation to get past SNI-based filtering. Rust · Tauri
Trace Linux machine observatory and control center: live stats, history, slowness diagnosis, plus control over processes, cgroups, systemd, Docker and GPU. Rust · Tauri · React
Cortex A private coding agent for VS Code that runs entirely on local Ollama models. JavaScript
DayZero · live Climate stress-testing engine that works out the month a real neighbourhood runs out of water, then finds the cheapest plan to buy those months back. Python · FastAPI · React · MapLibre
Neural World Model A world model that learns physics from a bouncing-ball sandbox and plans by imagining branching futures before it acts. PyTorch

🧰 What I reach for

When the job is I reach for
Native, fast, close to the OS Rust · Tauri
Models, training, experiments Python · PyTorch · OpenCV · scikit-learn
LLM apps and agents Ollama · LangChain · RAG · MCP
APIs and backends FastAPI · Node.js · PostgreSQL · MongoDB
Interfaces TypeScript · React · Next.js · Three.js
Shipping it Docker · GitHub Actions · Git
Where it all runs Arch Linux · Hyprland

Still in training: Solidity · Web3 · Distributed Systems · System Design


🎯 Current Arc

  • 🧠 Shipping local-first AI agents that run on consumer hardware
  • 🦀 Going deeper on Rust for systems and desktop tooling
  • 🔭 Experimenting with world models and multi-modal AI
  • 🌍 Looking for open source projects to contribute to

📊 Season Stats

GitHub stats Top languages

Contribution streak

Contribution snake


"Let's start from here. From zero." — Rem, Re:Zero

Pinned Loading

  1. Zenith Zenith Public

    Self-hosted, local-first AI workspace on your own Ollama models — chat, a Claude-Code-class coding agent, deep research, long-term memory, image gen, and voice. No cloud, no accounts, nothing leave…

    Python 3

  2. Zenith-AI Zenith-AI Public

    Zenith AI — a small language model built entirely from scratch: BPE tokenizer, RoPE/GQA transformer, training loop, KV-cache inference, and CLI, trained end-to-end on a single 4GB GPU.

    Python

  3. Cortex Cortex Public

    A local, private coding agent for VS Code — like Cline, but powered entirely by your own Ollama models. No API keys, no cloud.

    JavaScript

  4. Trace Trace Public

    Linux machine observatory + control center — live system stats, history, slowness diagnosis, project scanning, and a privileged control layer (processes, cgroups, systemd, Docker, GPU), built with …

    Rust