Skip to content
View zohaibus's full-sized avatar

Block or report zohaibus

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
zohaibus/README.md

Zohaib Usmani

Principal Engineer  ·  Silicon, Hybrid AI Orchestration & Sovereign Compute

From the antenna to the AI model, and the power rail to the thermal floor.
15+ years across the full edge-compute stack: RF, analog, baseband, processors, accelerators, and the AI that runs on top.

Open the LocalOffice Hub

LinkedIn MIT Local-first


I've spent 15+ years shipping production silicon and AI infrastructure that runs in billions of devices (RF, modem, NPU). Now I also build local-first tools for the post-cloud transition: software that runs entirely on your device, with no cloud, no account, and no telemetry.

The thesis: AI inference is a physical routing problem. Cloud-only AI is thermally and economically unsustainable. A fully isolated edge device hits memory-bandwidth walls. The future is a multi-tier hybrid split.

Tier Runs Handles
Edge / local box Small models, instant response, tight power envelope Most routine agentic work: routing, extraction, synthesis
On-prem / frontier cloud Massive memory, custom silicon The genuinely hard reasoning

I build the local-first application layer and the deterministic kernels designed to survive that shift. One rule runs through all of it: AI advises, deterministic code decides.


What I'm Building

A suite of single-file, local-first tools on one open JSON protocol (localoffice/v1). Local models draft; lightweight deterministic kernels verify. No install, no cloud, your data stays on your disk.

Open the Hub

The Hub opens the whole suite, or launch any tool directly:

Tool What it does Launch
LocalSheets Spreadsheet with a local AI panel ▶ Open
localDeck Single-file slide editor ▶ Open
localDoc Doc writer with a compliance linter ▶ Open
localCards Spaced-repetition flashcards ▶ Open
localMindMap Canvas mind-mapping ▶ Open
localMark Image sanitizer (strips metadata) ▶ Open
localCheck SOP / QA runbook with a gated sign-off ▶ Open
LocalPlan Single-file planner ▶ Open

A presentation editor in a single HTML file. Generate, lay out, and ship decks locally, no cloud suites involved.

Launch DeckBuilder

A human-first protocol for stateful LLM context and memory. Plain Markdown, Git-versionable, model-agnostic, fully offline.

All MIT licensed. Fork them, inspect the network logs, and own your compute.


Background

  • University of Washington  ·  BS/MS Electrical Engineering
  • Stanford  ·  AI Professional Program
  • MIT  ·  Product Management Professional Program
  • Angel investor  ·  early-stage Physical AI, deep tech, and edge infrastructure

What I'm Thinking About

  • Sovereign edge AI, from silicon to application
  • On-device LLMs and VLMs for robotics and physical execution
  • Hardware/software co-design at the power and thermal floor
  • Local-first software as a permanent category

Local-first. No cloud. No accounts. Your data stays on your disk.

Pinned Loading

  1. localOffice localOffice Public

    Single-file, local-first office tools: fully offline, zero dependencies, built on one rule: AI advises, deterministic math decides.

    HTML 1

  2. context-protocol context-protocol Public

    Sovereign Context Protocol - A human-first workflow for managing LLM memory. Not an MCP server.

    Python

  3. deckbuilder deckbuilder Public

    A local-first presentation editor in a single HTML file

    HTML 7

  4. localsheets localsheets Public

    A single-file, local-first, multi-sheet spreadsheet

    HTML 4