<?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: Jianqiang wang</title>
    <description>The latest articles on DEV Community by Jianqiang wang (@jianqiang_wang_06b88ac60c).</description>
    <link>https://dev.to/jianqiang_wang_06b88ac60c</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%2F4251690%2F703ad118-d012-4749-a607-35f5f679baae.jpg</url>
      <title>DEV Community: Jianqiang wang</title>
      <link>https://dev.to/jianqiang_wang_06b88ac60c</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/jianqiang_wang_06b88ac60c"/>
    <language>en</language>
    <item>
      <title>Our Photo-Retouching Agent Passed QA. The Edit Was Still Unfinished.</title>
      <dc:creator>Jianqiang wang</dc:creator>
      <pubDate>Fri, 20 Nov 2026 13:58:36 +0000</pubDate>
      <link>https://dev.to/jianqiang_wang_06b88ac60c/our-photo-retouching-agent-passed-qa-the-edit-was-still-unfinished-p0d</link>
      <guid>https://dev.to/jianqiang_wang_06b88ac60c/our-photo-retouching-agent-passed-qa-the-edit-was-still-unfinished-p0d</guid>
      <description>&lt;p&gt;&lt;em&gt;By Jay Wang / Curify. These notes describe our internal retouching research and workflow development through November 12, 2026. They do not establish a production-quality autonomous retoucher. AI assisted with drafting; the observations below are grounded in our engineering records.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Our retouching pipeline could produce an image, pass technical checks, and still leave requested work unfinished.&lt;/p&gt;

&lt;p&gt;In an October development audit, outputs that passed QA still included pending lighting work or no applied local repairs. We kept them marked as requiring review. The problem was not that the pipeline had secretly shipped them: it was that a green technical check answered a narrower question than “did we complete the edit?”&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Successful execution, preservation, and task completion need separate evidence.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;A request such as “reduce the shine, keep the skin texture, and leave the dress alone” contains several obligations: locate the issue, choose a scoped operation, apply an appropriate amount, and verify both the change and what should stay unchanged. A more attractive image can still fail that request.&lt;/p&gt;

&lt;p&gt;That is why we are approaching the agent through three sources: before/after pairs, professional knowledge, and recordings of the editing process. Each answers a different question. None can replace checking the result against the brief.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Before/after pairs describe a destination, not the route
&lt;/h2&gt;

&lt;p&gt;A pair can show the intended difference: a reduced color cast, a cleaner hair outline, or more even illumination. It rarely tells us which operation produced that difference—or which changes the client actually requested.&lt;/p&gt;

&lt;p&gt;Before using a pair, we need to know whether the images show the same source frame, whether their geometry matches, and whether resizing, cropping, compression, or watermarks affect the comparison. We also need the edit brief. Otherwise, “be more like the after image” can encourage unrelated changes.&lt;/p&gt;

&lt;p&gt;This distinction matters in our own collection. A reference useful for visual discussion is not automatically eligible for pixel-level evaluation. Multiple crops from one photograph are not independent examples. Images used to develop the workflow cannot later be presented as an unseen test set.&lt;/p&gt;

&lt;p&gt;A useful record connects the images to the task:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight json"&gt;&lt;code&gt;&lt;span class="p"&gt;{&lt;/span&gt;
  &lt;span class="nl"&gt;"edit_intent"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"reduce a localized highlight on the cheek"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"preserve"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"skin texture"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"expression"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"jewelry"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"dress detail"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"reference_role"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"lighting direction, not identity replacement"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
  &lt;span class="nl"&gt;"comparison_status"&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="s2"&gt;"requires alignment and region review"&lt;/span&gt;
&lt;span class="p"&gt;}&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;This is an illustrative schema, not a client record or an implemented API contract.&lt;/p&gt;

&lt;p&gt;We also compare against doing nothing. In one exposed development replay, the agent's average reference error was worse than the unchanged input. That does not settle every question about visual quality, but it prevents us from treating “the agent produced an image” as evidence of improvement.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Domain knowledge needs to become an executable decision
&lt;/h2&gt;

&lt;p&gt;Professional advice often arrives as a short phrase: preserve texture, separate lighting from blemish repair, or avoid turning every white surface into the same neutral white.&lt;/p&gt;

&lt;p&gt;An agent needs more than the phrase. It needs conditions for using a technique, a target region, an allowed strength, protected properties, and a way to decide whether the result helped.&lt;/p&gt;

&lt;p&gt;Consider “brighten the skin slightly.” The face, hands, and another person in the photograph may require different treatment. Brightening a whole rectangle can also change hair, fabric, or reflected light. A numerical adjustment applied successfully does not establish that it was appropriate.&lt;/p&gt;

&lt;p&gt;We therefore treat knowledge as a candidate editing policy that must be tested under a defined task. A proposed policy might separate illumination correction from local defect repair and require a texture check after either operation. It is a hypothesis until the comparison supports it.&lt;/p&gt;

&lt;p&gt;The useful evaluation separates three questions:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Question&lt;/th&gt;
&lt;th&gt;What we inspect&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Was the requested edit completed?&lt;/td&gt;
&lt;td&gt;The defect or region named in the brief&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Were protected details preserved?&lt;/td&gt;
&lt;td&gt;Identity, product structure, texture, composition, and excluded regions&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Does the user prefer this result?&lt;/td&gt;
&lt;td&gt;A review against the same brief, allowing rejection of every candidate&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;These are different outcomes. Passing a preservation check does not mean the blemish was removed. A user preferring a warmer look does not make a cooler expert reference the universal target.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Editing recordings expose decisions that pairs cannot
&lt;/h2&gt;

&lt;p&gt;A screen recording can show tool selection, masks, operation order, diagnostic views, and changes of direction. It can explain why a retoucher worked on illumination before texture, or why a local adjustment was reduced after viewing the whole image.&lt;/p&gt;

&lt;p&gt;But a recording is not a clean action trace. A cursor can move without changing pixels. A layer can be hidden. A diagnostic view can make the image look dramatically different without becoming part of the export. The recording can also begin after earlier edits have already happened.&lt;/p&gt;

&lt;p&gt;In our tutorial review, an automatic transcript rendered a brush-opacity value as 15%, while the visible interface showed 5%. Later frames also showed additional texture work. Attributing the final image entirely to the earlier curves adjustment would have been incorrect.&lt;/p&gt;

&lt;p&gt;That changed how we treat recordings: retain timestamps, inspect the actual interface, distinguish opacity from flow, and record what is visible versus what is inferred. A Photoshop brush setting is not automatically the correct parameter for a different editing tool.&lt;/p&gt;

&lt;p&gt;Our current recording work includes sampled frames and transcript drafts. It is not complete reconstruction of professional actions, nor evidence that we trained a model to reproduce the workflow. Publicly viewable tutorials also do not automatically grant training or redistribution rights; no tutorial footage or client photographs are included here.&lt;/p&gt;

&lt;h2&gt;
  
  
  Connecting the three sources to an agent
&lt;/h2&gt;

&lt;p&gt;The architecture we are working toward is:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Brief + reference role + protected details
                    |
           Select an editing policy
                    |
           Propose a scoped operation
                    |
           Execute and retain a version
                    |
       Check completion AND preservation
                    |
       Accept, revise, revert, or request review
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Pairs help describe the intended outcome. Knowledge helps propose an operation. Recordings help explain sequencing and intermediate decisions. Verification determines whether that particular operation was useful on that particular image.&lt;/p&gt;

&lt;p&gt;Here, “learning” primarily means improving the agent's planning, tool use, and evaluation workflow. We are not claiming that combining these resources has already trained new model weights or delivered a general improvement.&lt;/p&gt;

&lt;p&gt;Our engineering records include local edits, protected regions, execution logs, and review states. The full policy-learning and feedback architecture is still a research direction. In particular, a pipeline stage marked “skipped” or “pending” must not silently count as completed work.&lt;/p&gt;

&lt;h2&gt;
  
  
  Two failure modes worth keeping in the evaluation
&lt;/h2&gt;

&lt;p&gt;First, a metric can miss the edit. In an archived hair-edit review, some operations changed pixels entirely outside the frozen scoring region. An unchanged score therefore did not mean that nothing happened. The metric was uninformative about those hair changes. We kept the historical score and investigated the region coverage separately.&lt;/p&gt;

&lt;p&gt;Second, a technically acceptable file can contain unfinished work. Our completion audit found outputs that passed technical checks but still had pending lighting work or no applied local repairs. They remained review-required. A valid file and preserved geometry are useful, but they are not a substitute for completing the brief.&lt;/p&gt;

&lt;p&gt;A minimal review gate makes the distinction explicit:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="c1"&gt;# Illustrative review logic, not our production implementation.&lt;/span&gt;
&lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;review_status&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;technical_ok&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;task_checks&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;preservation_checks&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;pending&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="n"&gt;technical_ok&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="s2"&gt;"technical_failure"&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;pending&lt;/span&gt; &lt;span class="ow"&gt;or&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="n"&gt;task_checks&lt;/span&gt; &lt;span class="ow"&gt;or&lt;/span&gt; &lt;span class="ow"&gt;not&lt;/span&gt; &lt;span class="n"&gt;preservation_checks&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="s2"&gt;"review_required"&lt;/span&gt;
    &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="nb"&gt;any&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;check&lt;/span&gt; &lt;span class="o"&gt;!=&lt;/span&gt; &lt;span class="s2"&gt;"pass"&lt;/span&gt; &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;check&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="n"&gt;task_checks&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="n"&gt;preservation_checks&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="s2"&gt;"review_required"&lt;/span&gt;
    &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="s2"&gt;"candidate_for_human_review"&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The brief must define the checks before execution. An empty checklist is missing evidence, not a pass. Skipped repairs remain pending; failed preservation checks require revision or rollback. Even the final state above is a candidate for review, not delivery approval.&lt;/p&gt;

&lt;p&gt;These observations suggest a practical evaluation: compare the same inputs and budgets, record the first result and every revision, include failures in the denominator, and measure new damage as well as improvements. Independent visual review matters because a more elaborate verification loop can still agree with its own mistakes.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we would test next
&lt;/h2&gt;

&lt;p&gt;The next useful comparison is a general agent with tools, the same agent with a professional editing policy, and that policy with visual verification. Keep the inputs, tools, protected details, and total budget comparable.&lt;/p&gt;

&lt;p&gt;We would report task completion, preservation regressions, human acceptance, and cost per accepted result separately. For repeated edits, feedback must refer to a specific image version: “return to this version and brighten the skin slightly” should not inherit changes from a rejected candidate.&lt;/p&gt;

&lt;p&gt;The goal is an agent that can explain and check a small edit, then stop or ask for help when the evidence is insufficient. A polished before/after image is useful for showing an outcome. It is not, by itself, a record of how the outcome was obtained or whether the agent satisfied the request.&lt;/p&gt;

&lt;p&gt;For a practical companion, our &lt;a href="https://www.curify-ai.com/resources/ai-photo-retouching-quality-checklist" rel="noopener noreferrer"&gt;photo-retouching QA checklist&lt;/a&gt; covers agreeing scope and reviewing details before delivery. A separate &lt;a href="https://www.curify-ai.com/blog/batch-product-image-quality-control" rel="noopener noreferrer"&gt;batch product-image QA case&lt;/a&gt; describes targeted revisions in a 96-image production process; it is a production example, not a validation of this agent research.&lt;/p&gt;

&lt;h2&gt;
  
  
  About Curify
&lt;/h2&gt;

&lt;p&gt;I'm building &lt;a href="https://www.curify-ai.com" rel="noopener noreferrer"&gt;Curify&lt;/a&gt;, where we work on AI-assisted visual production and tools for creators and teams. Photo retouching is one part of that work: turning a brief into edits that can be reviewed, with clear limits on what changed and what remains uncertain. For a commercial batch, our &lt;a href="https://www.curify-ai.com/tools/product-photo-retouching-service" rel="noopener noreferrer"&gt;product photo retouching service&lt;/a&gt; starts with representative images and an agreed scope.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;How do you evaluate whether a requested edit succeeded without changing protected details?&lt;/strong&gt; If you build editing agents, I would be especially interested in cases where your automated checks passed but a human reviewer still rejected the result.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>computervision</category>
      <category>machinelearning</category>
      <category>programming</category>
    </item>
    <item>
      <title>Fluent English is not evidence of ASL understanding: lessons from building a video translator</title>
      <dc:creator>Jianqiang wang</dc:creator>
      <pubDate>Wed, 18 Nov 2026 10:49:19 +0000</pubDate>
      <link>https://dev.to/jianqiang_wang_06b88ac60c/fluent-english-is-not-evidence-of-asl-understanding-lessons-from-building-a-video-translator-5ca6</link>
      <guid>https://dev.to/jianqiang_wang_06b88ac60c/fluent-english-is-not-evidence-of-asl-understanding-lessons-from-building-a-video-translator-5ca6</guid>
      <description>&lt;p&gt;&lt;em&gt;By Jay Wang / Curify. Adapted from our engineering account of experiments recorded through October 12, 2026. This describes historical research, not a new evaluation of today's production model.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;A fluent English sentence can make an ASL translation demo look convincing. It can also hide the fact that the model did not understand the signing.&lt;/p&gt;

&lt;p&gt;While building Curify's ASL video translator, we encountered several ways an apparently successful result could mislead us. The useful lesson was to make each prediction inspectable, and to keep the claim narrower than the evidence.&lt;/p&gt;

&lt;h2&gt;
  
  
  First check whether the answer is already in the input
&lt;/h2&gt;

&lt;p&gt;An early demo appeared to translate a signed clip well. Inspecting it showed that the model was reading captions burned into the frames. We retracted that result as evidence of sign-language understanding.&lt;/p&gt;

&lt;p&gt;Video input does not make an evaluation independent. Visible text can leak the answer. So can a prompt or a reference translation created without a documented annotation method.&lt;/p&gt;

&lt;p&gt;We separated diagnostic clips from independent evaluation material. A diagnostic example is useful for finding a problem. It does not automatically belong in an accuracy report.&lt;/p&gt;

&lt;h2&gt;
  
  
  Inspect the input from the saved run
&lt;/h2&gt;

&lt;p&gt;Our frame inspector aligns sampled frames, batch boundaries, reference cues and saved predictions on a timeline. That lets us ask a concrete question: did the recognizer receive enough of the visual event to support the English it produced?&lt;/p&gt;

&lt;p&gt;The inspector must match the actual run. Resampling the same clip with a newer configuration would show different evidence. We preserve the sampling configuration alongside predictions; a historical 96-frame sheet is not interchangeable with later duration-dependent sampling.&lt;/p&gt;

&lt;p&gt;A minimal inspection record should connect:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;The input identifier and permission scope.&lt;/li&gt;
&lt;li&gt;Model, prompt and preprocessing versions.&lt;/li&gt;
&lt;li&gt;Sampled timestamps and batch boundaries.&lt;/li&gt;
&lt;li&gt;The saved prediction and the provenance of its reference.&lt;/li&gt;
&lt;li&gt;A review decision, with the reason and unresolved questions.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The first four items support reproducibility. The final item still needs appropriate ASL expertise.&lt;/p&gt;

&lt;h2&gt;
  
  
  More frames and better text metrics were not sufficient
&lt;/h2&gt;

&lt;p&gt;We tried denser sampling, uniform sampling and overlapping batches with shared context. In our 12-clip diagnostic study, none of these frame-based variants recovered correct meaning. That is a finding about this setup and sample, not evidence that temporal modeling cannot help.&lt;/p&gt;

&lt;p&gt;In one batching comparison, word error rate improved while character overlap fell and outputs became shorter. A single metric could have rewarded the system for saying less. Lowering generation temperature can likewise improve repeatability without improving interpretation.&lt;/p&gt;

&lt;p&gt;Our controlled training work raised another distinction. Increasing training data at fixed initialization and compute did not produce a sustained BLEU gain on a frozen 147-clip evaluation. A separate representation experiment improved text-overlap metrics, but differences in pretraining and preprocessing limited the conclusions we could draw. Those numbers were not customer-video accuracy.&lt;/p&gt;

&lt;p&gt;The full &lt;a href="https://www.curify-ai.com/blog/building-asl-video-translator" rel="noopener noreferrer"&gt;engineering account&lt;/a&gt; includes the experiment table, sample sizes and limitations. These are internally recorded findings, not an independently replicated benchmark.&lt;/p&gt;

&lt;h2&gt;
  
  
  The review question is whether meaning survived
&lt;/h2&gt;

&lt;p&gt;We want expert feedback on an evaluation rubric that distinguishes:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;Omissions, additions and changes in who did what to whom.&lt;/li&gt;
&lt;li&gt;Fingerspelling, names and numbers supported by visible evidence versus guesses.&lt;/li&gt;
&lt;li&gt;Nonmanual cues and spatial references that affect interpretation.&lt;/li&gt;
&lt;li&gt;Legitimate signing variation, which should not automatically be treated as an error.&lt;/li&gt;
&lt;li&gt;Cases where the output should express uncertainty or ask for clearer input.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;This is a proposed review framework, not a validated ASL scoring instrument. Agreement about rubric anchors should come before claims about accuracy. A research dataset result also needs a separate check on the intended input distribution before it can support a product claim.&lt;/p&gt;

&lt;h2&gt;
  
  
  What we would like to learn next
&lt;/h2&gt;

&lt;p&gt;For an ASL educator, researcher or fluent reviewer: which of these dimensions would you change, and what would a bounded first review need to establish?&lt;/p&gt;

&lt;p&gt;We are looking for a discussion of the evaluation method. We are not asking readers to endorse the tool, share student data, or send private signer videos. Any later sample review would need an agreed scope and permission-cleared material.&lt;/p&gt;

&lt;p&gt;You can read the &lt;a href="https://www.curify-ai.com/blog/building-asl-video-translator" rel="noopener noreferrer"&gt;original methods and limitations&lt;/a&gt;, or contact Jay at &lt;a href="mailto:team@curify-ai.com"&gt;team@curify-ai.com&lt;/a&gt;. The &lt;a href="https://www.curify-ai.com/tools/asl-video-translator" rel="noopener noreferrer"&gt;ASL video-to-English tool&lt;/a&gt; is a separate product surface: its current price and limitations apply, and a successful export does not establish translation correctness.&lt;/p&gt;

&lt;p&gt;Research context: &lt;a href="https://how2sign.github.io/" rel="noopener noreferrer"&gt;How2Sign&lt;/a&gt; and &lt;a href="https://github.com/ZechengLi19/Uni-Sign" rel="noopener noreferrer"&gt;Uni-Sign&lt;/a&gt;. We credit the upstream work; our integration experiments are not a claim to have reproduced every upstream result.&lt;/p&gt;

</description>
      <category>machinelearning</category>
      <category>computervision</category>
      <category>a11y</category>
    </item>
  </channel>
</rss>
