Skip to content

Get in touch

Send me a message, or book a time to talk.

tengariya@gmail.com

Tho8

A workspace for your notes that shows why it suggests what it does.

Tho8 at 1440 wide in its dark theme: the notes list on the left, a note with Ask open, using the full note, on the question "Why does an explanation raise trust?", showing 5 of 6 matching sentences from the note, and the Links panel on the right with 4 suggestions and 12 linked notes.
The workspace: notes on the left, the note with Ask open in the middle, and the Links panel on the right.
Detail of Ask, using the full note: the question "Why does an explanation raise trust?", Matching sentences 5 of 6, three sentences quoted from the note with the matching words lit, and Insert below, Create note, Copy and Discard under them.
Detail. Ask quotes the sentences in the note that match the question.
Tho8 on a phone in its dark theme: Ask, using the full note, on "What is prospect theory?". The card reads Nothing in this note holds enough of the question, lists the words it looked for, prospect and theory, and offers Ask all notes.
When the note holds no answer, Ask says so and shows the words it looked for.

People lose their ideas later, when they try to get back to them. So I designed a workspace for that. You save a thought with a shortcut, each suggestion says why it’s there, and answers quote your own notes.

Role
Product design, visual design and the design system. I led the product and the interface.
Team
A two-person team called Luhmann's Quorum, with Ruhi Vadsariya, advised by Prof. Sang-Hwan Kim. Ruhi and I did the research, ideation, evaluation and testing together, and I did the design.
Date
2025 to 2026. A two-semester capstone for my master’s in Human-Centered Design and Engineering at the University of Michigan.
Outcome
A working prototype that was never released, so it has no users. We ran two rounds of moderated usability testing on the first build. The screens here show the current build with a demo vault of made-up notes.

Ideas get lost when people go back for them.

In our interviews the same loop kept breaking. People wrote an idea down and then couldn't find it, couldn't link it to anything, or didn't trust the AI that was trying to help.

Four loops in one workspace.

There are four loops that feed into each other, so when one of them breaks, the whole thing feels broken.

  1. CaptureSaving a thought with a shortcut
  2. ConnectLinking notes by what they mean
  3. RetrieveFinding an idea in thirty seconds or less
  4. ResurfaceBringing an idea back when it helps

Every suggestion shows its reason.

Each suggestion comes with a small label saying why it was made: a tag both notes share, a note it came through, or similar content. Opening the label shows the evidence, the note both ideas connect to, and the similarity score. Link and Dismiss are always visible.

When we asked people what would make them trust a suggestion, they ranked the kinds of evidence in a consistent order.

  1. A shared sourceBoth notes cite the same paper
  2. A two-hop pathA to B to C, a route you can follow
  3. A specific termA distinctive phrase both notes use

Earns a click

  1. Tags or recencyWeak evidence, and easy to distrust

Dismissed

Tho8 at 1440 wide in its dark theme: a note open in the middle and, in the Links panel, Suggested by Tho8, 4. The first suggestion, Automation gets discounted faster than a person after one bad forecast, has its via chip open on a card: Evidence, a quoted passage, Both connected to People trust an algorithm less after watching it make one mistake, and Similarity 0.33 (suggestions start at 0.20).
A suggestion with its reason open.
Detail of the Links panel: Suggested by Tho8, 4, and the first suggestion, Automation gets discounted faster than a person after one bad forecast, with its via chip open on a card: Evidence, a quoted passage in italics, Both connected to People trust an algorithm less after watching it make one mistake, Similarity 0.33 (suggestions start at 0.20), and Link as.
Detail. Opening the reason shows the evidence, the note both connect to, and the score.

Help that quotes your own notes.

Tho8’s help sits on the editor’s format bar, and Cmd+J opens it too. It has five modes: Ask, Summarize, From linked notes, Draft and Improve.

Tho8 doesn’t write anything new. Every result is a sentence from your own notes with its source, an edit, or a count. You see it first, choose what to do, and can undo it.

When none of your notes holds the answer, it says so and shows the words it looked for. If you misspell a word, it reads it as the word in your notes and tells you it did.

Detail of Draft, using all 48 notes, on "a memo on points and defaults": an Outline of 5, each a note title with sentences quoted from it in italics, and under them Insert below, Create note, Copy, Discard and Adds links to 3 notes.
Detail. Draft builds an outline from sentences already in your notes, each under the note it came from.
Draft on a phone in the dark theme, using all 48 notes, on "a memo on points and defaults": an Outline of 5, the first note, Defaults are the strongest nudge, with two quoted sentences, then Insert below, Create note, Copy, Discard and Adds links to 3 notes.
Draft builds an outline from sentences already in your notes.

A graph that opens on the note you are in.

Links in and out sit on either side of your note, with Tho8’s suggestions underneath. You can look one or two steps out. Switch to the whole vault and it shows everything, tags and folders included, and lists the notes that have no links.

Every line tells you why two notes are linked, either with the sentence that makes the link or with Tho8’s reason. You can also move around the graph from the keyboard.

Tho8 at 1440 wide in its dark theme: the graph in a tab, set to This note and 1 step, with Explanations raise trust in algorithmic advice boxed in the middle, the notes it links to drawn around it with arrows, four of Tho8’s suggestions on dashed lines below it, a key for Note, Link and Suggested by Tho8, and the Links panel on the right.
The graph, opened on the note you are in.
Detail of the graph: Explanations raise trust in algorithmic advice boxed in the middle, the notes it links to around it with arrows, and four of Tho8’s suggestions below it on dashed lines, each marked with a link icon.
Detail. The note you are in, its links around it, and Tho8’s suggestions dashed below.
The graph on a phone in the dark theme, set to This note and 1 step: Explanations raise trust in algorithmic advice boxed in the middle, the notes it links to above and around it, and four of Tho8’s suggestions dashed below.
Detail. The graph on a phone, opened on the note you are in.

Finding notes and saving thoughts from the keyboard.

Cmd+K opens a search at once. It finds notes, actions and matching lines inside notes, and if nothing matches it offers to create a note with that name.

Quick capture, with Cmd+Shift+Space, saves a thought to the Inbox along with the name of the note you were reading. Templates show a live preview as you pick one.

Tho8 at 1440 wide in its dark theme with the command palette open over a note: Search notes or type a command, five Recently opened notes with their folders, Views (Recent, Favorites, Worth another look and Trash), Actions starting with New note, and the keys along the bottom: arrows to move, Return to open, Cmd+Return to open in a new tab, Esc to close.
The Cmd+K search.
Detail of the command palette: Search notes or type a command, five Recently opened notes with their folders, Views (Recent, Favorites, Worth another look and Trash), Actions starting with New note, and the keys: arrows to move, Return to open, Cmd+Return to open in a new tab, Esc to close.
Detail. The Cmd+K search.
The command palette on a phone in the dark theme: Search notes or type a command, five Recently opened notes with their folders, and Views: Recent, Favorites, Worth another look and Trash.
Detail. The Cmd+K search on a phone.

What testing changed in the first build.

Ruhi and I ran two rounds of moderated usability testing on the first build, with six people. That’s a small sample, and nobody has used it outside the testing.

People trusted a suggestion once they could see why it was there, and ignored it when they couldn't, even if it was a good one.

  1. The graph overwhelmed peopleIt opens focused on the note you clicked, with a clear way back.
  2. The AI’s scope was a guessEvery answer is labelled with what it is based on.
  3. Nobody knew if a note had savedYou can see when a note has saved.
  4. Trust went up when people could see the reasoningShowing why became part of the main view.

What I took from it.

I think people trust an AI feature more when it shows the reason, the source and a score, and makes it easy to say no.

Tho8 at 1440 wide in its dark theme with Settings open on the Tho8 section: Use Tho8 on, Suggestions set to Light assist, Show why always on, Bring a note back after 30 days with 9 notes untouched this long, the 5 oldest show in Worth another look, and Keep Tho8 out of, with folders to switch off.
The Tho8 settings, where Show why is always on: "Every suggestion shows what it rests on."
Detail of the Tho8 settings: Use Tho8 on, Suggestions set to Light assist, Show why always on, every suggestion shows what it rests on: shared tags, a shared neighbour or similar content, and Bring a note back after 30 days.
Detail. Show why is always on: "Every suggestion shows what it rests on."
The Tho8 settings on a phone in the dark theme: Use Tho8 on, Suggestions set to Light assist, Show why always on, and Bring a note back after 30 days.
Detail. Show why is always on: "Every suggestion shows what it rests on."

All work