<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Lukasz Snopkiewicz</title>
    <description>The latest articles on DEV Community by Lukasz Snopkiewicz (@lukasz_snopkiewicz_0d1bab).</description>
    <link>https://dev.to/lukasz_snopkiewicz_0d1bab</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F2999212%2F42f6daf5-a09e-4361-8d6a-40e703e808c6.png</url>
      <title>DEV Community: Lukasz Snopkiewicz</title>
      <link>https://dev.to/lukasz_snopkiewicz_0d1bab</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/lukasz_snopkiewicz_0d1bab"/>
    <language>en</language>
    <item>
      <title>Tidal Loom: let your breathing draw a ribbon of ink</title>
      <dc:creator>Lukasz Snopkiewicz</dc:creator>
      <pubDate>Fri, 06 Nov 2026 22:47:16 +0000</pubDate>
      <link>https://dev.to/lukasz_snopkiewicz_0d1bab/tidal-loom-let-your-breathing-draw-a-ribbon-of-ink-1l7d</link>
      <guid>https://dev.to/lukasz_snopkiewicz_0d1bab/tidal-loom-let-your-breathing-draw-a-ribbon-of-ink-1l7d</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/presage-2026-11-04"&gt;Presage SmartSpectra SDK Challenge&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Tidal Loom turns natural chest movement into layered ribbons of ink.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;Choose terracotta, sage or ocean, sit comfortably, and let a small canvas respond to you. There is no target breathing rate, score or assessment. The interaction is simply movement becoming a mark.&lt;/p&gt;

&lt;p&gt;A labelled illustration mode lets anyone explore the artwork before choosing camera processing. Live mode uses SmartSpectra’s chest-breathing trace. An artwork-only recorder can export a short video without including camera images or audio.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://soundcake.github.io/tidal-loom-presage-20261106/" rel="noopener noreferrer"&gt;Watch the 51-second live SmartSpectra demonstration&lt;/a&gt;. Natural chest movement controls the ribbons; the recording includes only the artwork and its live signal label.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffusdjoy2d75o1hytmglz.png" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Ffusdjoy2d75o1hytmglz.png" alt="Tidal Loom responding to a live SmartSpectra chest-motion signal" width="800" height="500"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://soundcake.github.io/tidal-loom-presage-20261106/web/index.html" rel="noopener noreferrer"&gt;Try the illustration in your browser&lt;/a&gt;. No account or camera is needed for that mode. To run it locally, clone the repository, run &lt;code&gt;node preview.mjs&lt;/code&gt; and open &lt;code&gt;http://127.0.0.1:52106&lt;/code&gt;.&lt;/p&gt;

&lt;p&gt;For live mode, follow the README’s desktop setup: &lt;code&gt;npm ci&lt;/code&gt;, &lt;code&gt;npm run build:live&lt;/code&gt;, then &lt;code&gt;npm run start:live&lt;/code&gt;. Use your own SmartSpectra API key, read the data notice, give consent and choose &lt;strong&gt;Start with my camera&lt;/strong&gt;. The app does not require its own account. Place your face and upper chest in view, under steady lighting, and breathe naturally.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/soundcake/tidal-loom-presage-20261106" rel="noopener noreferrer"&gt;Source and setup instructions on GitHub&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The app uses Electron, Canvas 2D and SmartSpectra 3.4.0. Eight automated tests cover the signal’s validation/freshness gates and the sensor lifecycle, including stopping while camera permission is still pending.&lt;/p&gt;

&lt;h2&gt;
  
  
  How It Responds to Human Signals
&lt;/h2&gt;

&lt;p&gt;The SDK requests &lt;code&gt;MetricType.CHEST_BREATHING&lt;/code&gt;. The app decodes &lt;code&gt;breathing.upperTrace&lt;/code&gt;, normalizes a short rolling window and smooths the resulting value. That value changes the ribbons’ amplitude, making the composition open and narrow with chest movement.&lt;/p&gt;

&lt;p&gt;When validation fails or the data becomes stale, live animation pauses. It requires fresh valid samples to resume. Late callbacks from an ended session are ignored. Synthetic illustration mode stays explicitly labelled, including in exported footage.&lt;/p&gt;

&lt;h2&gt;
  
  
  Prize Categories
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Most Creative or Delightful Use of Human Signals.&lt;/strong&gt; Tidal Loom uses the signal as an art material: a direct, visible interaction without turning it into a performance target.&lt;/p&gt;

&lt;h2&gt;
  
  
  Consent and Data Handling
&lt;/h2&gt;

&lt;p&gt;Camera mode is for adults aged 18 or older and requires an explicit consent checkbox and Start. Stop clears the local motion window and ends capture. Hiding or closing the app also ends capture.&lt;/p&gt;

&lt;p&gt;Camera images remain on the device. Tidal Loom holds at most 180 movement samples in memory and does not maintain an application database or add analytics.&lt;/p&gt;

&lt;p&gt;Optional SDK telemetry and accumulated output are disabled. The &lt;a href="https://presagetech.com/legal/privacy-policy" rel="noopener noreferrer"&gt;SmartSpectra SDK privacy policy&lt;/a&gt; nevertheless describes automatic transmission of derived measurements to Presage and Anthropic for Insights; that processing cannot currently be disabled. Authentication and usage records also remain. Tidal Loom does not add a prompt or display the resulting Insights.&lt;/p&gt;

&lt;p&gt;Presage describes retaining Insights while the developer account is active and normally deleting them within 60 days after account closure; other records have separate retention periods. Stopping the app clears local samples, not server records. The linked SDK policy explains deletion requests and the full retention schedule.&lt;/p&gt;

&lt;p&gt;Recording is optional. It captures only the artwork and signal-source label, with no camera frames, audio or API key. The video stays in memory until replaced or the app closes, unless explicitly downloaded. The recorder uploads nothing; downloaded files can be deleted locally.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>presagechallenge</category>
    </item>
    <item>
      <title>Lantern Archive: find a remembered story in its original words</title>
      <dc:creator>Lukasz Snopkiewicz</dc:creator>
      <pubDate>Thu, 29 Oct 2026 22:36:41 +0000</pubDate>
      <link>https://dev.to/lukasz_snopkiewicz_0d1bab/lantern-archive-find-a-remembered-story-in-its-original-words-324o</link>
      <guid>https://dev.to/lukasz_snopkiewicz_0d1bab/lantern-archive-find-a-remembered-story-in-its-original-words-324o</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week4-2026-10-26"&gt;Hacktoberfest Open-Source AI Challenge: Ghost in the Machine&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;Lantern Archive is a small offline search tool for letters, journals and notes.&lt;br&gt;
You can remember the meaning of a story without remembering the words that&lt;br&gt;
would find it. Lantern uses an open embedding model to connect a question to&lt;br&gt;
the relevant passage, then shows the original words and where to find them.&lt;/p&gt;

&lt;p&gt;In the included fictional archive, “Who fixed the broken timepiece?” finds&lt;br&gt;
Elsie's repair of the station clock. “What gave the homemade loaf its citrus&lt;br&gt;
fragrance?” finds the letter about orange peel rubbed into sugar.&lt;/p&gt;

&lt;p&gt;The “ghost” is a trace already present in the source: a small detail waiting to&lt;br&gt;
be found again. Every result includes its file and line numbers so the reader&lt;br&gt;
can return to the full story.&lt;/p&gt;
&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/soundcake/lantern-archive-hacktoberfest-20261029/blob/main/demo/walkthrough.mp4" rel="noopener noreferrer"&gt;Watch the 48-second walkthrough&lt;/a&gt; · &lt;a href="https://github.com/soundcake/lantern-archive-hacktoberfest-20261029/blob/main/demo/transcript.txt" rel="noopener noreferrer"&gt;Read the exact search results&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The walkthrough shows three searches against four fictional letters. Its&lt;br&gt;
captions come from an actual local inference run. The saved JSON and transcript&lt;br&gt;
are included alongside the video in the repository's &lt;code&gt;demo/&lt;/code&gt; folder.&lt;/p&gt;

&lt;p&gt;To run the same demonstration:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;npm ci
npm run setup:model
npm run demo
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Setup downloads the pinned model once. The demonstration then disables fetch&lt;br&gt;
requests while the model loads, indexes the archive and searches it. This run&lt;br&gt;
recorded zero attempted requests.&lt;/p&gt;
&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;&lt;a href="https://github.com/soundcake/lantern-archive-hacktoberfest-20261029" rel="noopener noreferrer"&gt;Source code and setup instructions&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The repository contains the CLI, source-indexing logic, a model setup script,&lt;br&gt;
fictional example letters, tests and the recorded demonstration. The app and&lt;br&gt;
examples use the MIT license.&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;node src/cli.mjs index examples/letters memories.lantern.json
node src/cli.mjs ask memories.lantern.json &lt;span class="s2"&gt;"Who fixed the broken timepiece?"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  How I Built It
&lt;/h2&gt;

&lt;p&gt;Lantern runs &lt;a href="https://huggingface.co/Xenova/all-MiniLM-L6-v2" rel="noopener noreferrer"&gt;Xenova's quantized all-MiniLM-L6-v2 model&lt;/a&gt;&lt;br&gt;
through &lt;a href="https://github.com/huggingface/transformers.js" rel="noopener noreferrer"&gt;Transformers.js&lt;/a&gt;.&lt;br&gt;
The model builds on the &lt;a href="https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2" rel="noopener noreferrer"&gt;Sentence Transformers all-MiniLM-L6-v2&lt;/a&gt;&lt;br&gt;
sentence-embedding work. The model weights and Transformers.js use Apache-2.0.&lt;br&gt;
The code pins both the model revision and the library version.&lt;/p&gt;

&lt;p&gt;Documents are split into passages with their original character offsets and&lt;br&gt;
line numbers. Each passage becomes a 384-dimensional vector. A question becomes&lt;br&gt;
another vector, and cosine similarity selects up to three passages above a&lt;br&gt;
threshold. Returned text comes directly from those source passages.&lt;/p&gt;

&lt;p&gt;Source SHA-256 hashes are checked before each search. If a document changes,&lt;br&gt;
Lantern asks for a rebuilt index, keeping quotations and line references tied&lt;br&gt;
to the current document. An explicit no-match result handles questions whose&lt;br&gt;
scores fall below the threshold.&lt;/p&gt;

&lt;p&gt;Nine core tests pass, covering quotation accuracy, line references, changed&lt;br&gt;
sources, passage splitting, ranking, no-match behavior and index handling. A&lt;br&gt;
separate real-model test passes four paraphrase searches with fetch disabled.&lt;br&gt;
The demonstration indexed twelve passages in 23 ms and answered its three&lt;br&gt;
queries in 1.6–3.8 ms each on macOS arm64; the saved report records the runtime&lt;br&gt;
and measurements for this small archive.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Does Open Innovation Matter?
&lt;/h2&gt;

&lt;p&gt;A personal archive can outlast the service used to create it. Open weights let&lt;br&gt;
Lantern search locally, without a hosted inference account or a subscription.&lt;br&gt;
The model, its revision and the retrieval code are available for inspection.&lt;br&gt;
Someone can repeat the same experiment, change the threshold or replace the&lt;br&gt;
model as their archive grows.&lt;/p&gt;

&lt;p&gt;That matters for the theme: keeping a story accessible includes keeping the&lt;br&gt;
tools needed to find it accessible. Lantern brings an old passage back into&lt;br&gt;
view, with its original wording intact.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
  </channel>
</rss>
