Han Tools

Han Tools Hannia Valera

I build things I wanted to exist.

Some are useful. Some are just for fun.

A few are both.

Hey, I’m Hannia Valera, a Software Engineer II at Microsoft working on developer experience, developer tooling, and AI-assisted software development. I build software that makes technology easier to use, whether that’s improving how developers work, creating intuitive product experiences, or turning complicated systems into tools people actually enjoy using.

Before Microsoft, I worked in software engineering at Apple, and I studied computer science at Northwestern University.

Han Tools is where my side projects, experiments, and random ideas end up. Some are useful developer tools. Others are games I wish existed. Mostly, I just like building things.

Featured project

Purrplex

A cozy cat logic puzzle

Put one cat in every row, column and color region, with no two cats touching, diagonals included.

  • Free. No ads, no tracking.
  • 100 levels, then unlimited generated puzzles. Every puzzle is verified to have exactly one solution.
  • Works offline. Keep the whole game as a single HTML file.
Play Purrplex Download Offline

A single HTML file that works without internet.

Also in Spanish: Jugar en español

How it works

Filed under
Game
Status
Live
Tags
  • logic puzzle
  • offline
  • puzzle generator
  • TypeScript

A real puzzle. Go on, try it.

1 of 4 cats placed.

One cat in every row, column and color region. No two cats touch, not even diagonally. The starter cat with the paw badge is already in the right place. Tap a square to place the selected mark. Arrow keys move, Space places, X marks, Backspace clears.

Featured project

Ryan

LLM Execution Explorer

Two runtimes, one prompt, different answers. Where did they split? Ryan runs the same conversation through two execution stacks and lines up what each one recorded, so you can see where the recorded evidence first differs. It is a browser-based tool for debugging differences between LLM inference stacks: it compares traces, and tokenization, chat templates and generation settings wherever both sides recorded them.

  • Runs in your browser with WebLLM and Transformers.js. Ollama has an adapter and a local bridge.
  • Compares the chat template, rendered prompt, tokenization, generation settings and decode steps where both sides recorded them, marks what a runtime does not expose as unknown, and shows where the outputs first diverge.
  • Imports traces from other tools and shares a comparison as a link. No accounts, and Ryan’s own code uploads no trace.

The first real run downloads about 480 MB of model files into your browser cache. WebGPU is recommended.

Open Ryan See an example comparison

How it works

Filed under
Developer tool
Status
Live
Tags
  • LLM tooling
  • WebGPU
  • WebLLM
  • Transformers.js
  • traces
Two token rows that match until the fifth tokenAn illustration of a Ryan comparison, with made-up data. Five stage markers (template, prompt, tokens, config, decode) are checked except the last, which shows a mismatch. Row A, from one runtime, and row B, from another, share the tokens The, sky, looks and blue, then split: because on one side, due on the other.templateprompttokensconfigdecodeARuntime ATheskylooksbluebecauseBRuntime BTheskylooksblueduetokens 1–4 matchfirst split
Illustration, not real output: two runtimes agree until the fifth token.

Private project

FreightTrust PayGuard

Freight transaction review and exception management

FreightTrust PayGuard is a private software project focused on helping logistics teams review freight documentation, identify inconsistencies, and organize issues requiring attention before critical operational or payment decisions.

Not available to the public: there is no demo, public beta or subscription.

Ask about this project

Opens my LinkedIn profile.

Filed under
Commercial software / Logistics
Status
Private development
Tags
  • logistics
  • freight
  • document review
  • exception management
Three overlapping sheets of paper with matching fields joined by linesAn abstract drawing. The sheets overlap, each a little lower than the one before, and each has a header bar, four bars of different lengths and a short bar at the foot. Curved lines, each with a small tick, join matching bars from one sheet to the next. One joining line is broken in the middle and circled in orange, and the bar it leads to is shorter than the one it comes from.
Abstract illustration, not a screenshot of the product.

The index

Projects

Things I’ve built, things I’m building, and ideas that somehow became real.

Things I’ve built

  • Purrplex

    A cozy cat logic puzzle

    • logic puzzle
    • offline
    • puzzle generator
    • TypeScript
    Filed under
    Game
    Status
    Live

    OpenHow it works

  • Ryan

    LLM Execution Explorer

    • LLM tooling
    • WebGPU
    • WebLLM
    • Transformers.js
    • traces
    Filed under
    Developer tool
    Status
    Live

    OpenHow it works

In private development

  • FreightTrust PayGuard

    Freight transaction review and exception management

    • logistics
    • freight
    • document review
    • exception management
    Filed under
    Commercial software / Logistics
    Status
    Private development

    Ask about this project

More on the way. This list grows as I finish things.

About

What I like building

I’m drawn to the layer of software most people never see: developer experience and tooling, the build and dev workflows that decide whether a day feels smooth or sticky, and the infrastructure that helps other engineers get their work done. I’m curious about AI-assisted software development and how it changes the way we build. Along the way I’ve also worked on compilers.

And I like small, independent experiments, the kind you can finish, ship and share with someone, like a cat puzzle or a trace explorer.