<?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: Jack Richer</title>
    <description>The latest articles on DEV Community by Jack Richer (@jricher).</description>
    <link>https://dev.to/jricher</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%2F4214830%2Fc9a0e7df-6252-4f98-8e5f-2115a638ec64.png</url>
      <title>DEV Community: Jack Richer</title>
      <link>https://dev.to/jricher</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/jricher"/>
    <language>en</language>
    <item>
      <title>How AI Visual Search Is Transforming Fashion Apps: Models, Implementation &amp; Costs</title>
      <dc:creator>Jack Richer</dc:creator>
      <pubDate>Wed, 18 Nov 2026 07:37:55 +0000</pubDate>
      <link>https://dev.to/jricher/how-ai-visual-search-is-transforming-fashion-apps-models-implementation-costs-4fpe</link>
      <guid>https://dev.to/jricher/how-ai-visual-search-is-transforming-fashion-apps-models-implementation-costs-4fpe</guid>
      <description>&lt;p&gt;Shoppers rarely type the exact product name. They type "black leather ankle boots" or upload a photo and expect the app to find something close. This post shows how to build that with open models, then looks at what it means for a fashion app budget, using Australia as the example market.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why it matters
&lt;/h2&gt;

&lt;p&gt;Online spending in Australia hit AUD 21.9 billion in April–June 2026, up 14% year on year, according to Australia Post. Fashion catalogues are huge and visual, and keyword search alone misses a lot. Cost guides for ecommerce apps, such as &lt;a href="https://www.techugo.com/blog/cost-to-develop-an-ecommerce-app-like-iconic-in-australia/" rel="noopener noreferrer"&gt;Techugo's breakdown for an app like THE ICONIC&lt;/a&gt;, list AI-based search, recommendations and visual discovery among the features that raise development effort.&lt;/p&gt;

&lt;h3&gt;
  
  
  The core idea: one embedding space for text and images
&lt;/h3&gt;

&lt;p&gt;A CLIP-style model maps images and text into the same vector space. Embed every product photo once, embed each query at request time, and return the nearest vectors.&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="n"&gt;python&lt;/span&gt;
&lt;span class="k"&gt;from&lt;/span&gt; &lt;span class="n"&gt;sentence_transformers&lt;/span&gt; &lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="nc"&gt;SentenceTransformer&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;util&lt;/span&gt;
&lt;span class="k"&gt;from&lt;/span&gt; &lt;span class="no"&gt;PIL&lt;/span&gt; &lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="nc"&gt;Image&lt;/span&gt;

&lt;span class="n"&gt;model&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nc"&gt;SentenceTransformer&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"clip-ViT-B-32"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

&lt;span class="c1"&gt;# 1. Offline: embed the catalogue images&lt;/span&gt;
&lt;span class="n"&gt;paths&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"boot_01.jpg"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"sneaker_02.jpg"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"dress_03.jpg"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
&lt;span class="n"&gt;img_emb&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;encode&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;
    &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="nc"&gt;Image&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nb"&gt;open&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;p&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;p&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="n"&gt;paths&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;convert_to_tensor&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="kc"&gt;True&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;normalize_embeddings&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="kc"&gt;True&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;)&lt;/span&gt;

&lt;span class="c1"&gt;# 2. Online: embed the shopper's text query&lt;/span&gt;
&lt;span class="n"&gt;query&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="s2"&gt;"black leather ankle boots"&lt;/span&gt;
&lt;span class="n"&gt;q_emb&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;model&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;encode&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;query&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;convert_to_tensor&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="kc"&gt;True&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;normalize_embeddings&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="kc"&gt;True&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

&lt;span class="c1"&gt;# 3. Retrieve the closest products&lt;/span&gt;
&lt;span class="n"&gt;hits&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;util&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;semantic_search&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;q_emb&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;img_emb&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;top_k&lt;/span&gt;&lt;span class="o"&gt;=&lt;/span&gt;&lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
&lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;h&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="n"&gt;hits&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="k"&gt;print&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;paths&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;h&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"corpus_id"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nb"&gt;round&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;h&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="s2"&gt;"score"&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="mi"&gt;3&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Passing an image as the query works the same way, which gives you "shop the look" for free.&lt;/p&gt;

&lt;h3&gt;
  
  
  Architecture in four parts
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;Ingestion job&lt;/strong&gt;: embeds new or changed product images on a schedule and writes vectors to an index (FAISS, pgvector or a managed vector database).&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Search API&lt;/strong&gt;: embeds the query, runs a nearest-neighbour lookup, then filters by size, stock and price.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Re-ranking&lt;/strong&gt;: blends similarity with business signals such as availability, margin or popularity.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;App layer&lt;/strong&gt;: a cross-platform client (Flutter or React Native) is usually enough, because the heavy lifting happens on the server.&lt;/p&gt;

&lt;h3&gt;
  
  
  Things that go wrong in practice
&lt;/h3&gt;

&lt;p&gt;&lt;strong&gt;Domain gap&lt;/strong&gt;: general-purpose models can confuse close categories such as "midi skirt" and "maxi skirt". Evaluate on your own catalogue and consider fine-tuning.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Stale vectors&lt;/strong&gt;: if images change and embeddings don't, results drift. Version your index.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Cold start&lt;/strong&gt;: new shoppers have no history, so fall back to visual similarity and trending items.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Measure it&lt;/strong&gt;: track recall@k on a small labelled query set, plus click-through and add-to-cart rate online.&lt;/p&gt;

&lt;h3&gt;
  
  
  What it means for the budget
&lt;/h3&gt;

&lt;p&gt;Reference ranges from the same cost guide, in AUD:&lt;/p&gt;

&lt;p&gt;Tier       -       Budget     -       Timeline&lt;br&gt;
MVP    -     30K – 150K      -      3–5 months&lt;br&gt;
Mid-level  -    150K – 250K      -      5–8 months&lt;br&gt;
Advanced   -    250K – 350K     -       8–12 months&lt;/p&gt;

&lt;p&gt;My suggestion is to ship plain search in the MVP, then add visual search once you have real queries to evaluate against. Beyond the build, plan for:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Hosting&lt;/strong&gt;: roughly AUD 300–3,000+ per month depending on traffic, with spikes in big sale seasons. Vector search and model inference add to this.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Maintenance&lt;/strong&gt;: about 15–20% of the initial build cost per year.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Compliance&lt;/strong&gt;: if you use shopper photos or behaviour data, the Privacy Act 1988 and the Australian Privacy Principles apply.&lt;/p&gt;

&lt;h3&gt;
  
  
  Takeaway
&lt;/h3&gt;

&lt;p&gt;Visual search is a small, well-understood pipeline: embed once, search fast, re-rank with business rules. Start with an open model, measure on your own catalogue, and let the data decide whether fine-tuning is worth the cost.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Resources&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.sbert.net/examples/sentence_transformer/applications/image-search/README.html" rel="noopener noreferrer"&gt;Sentence Transformers: image search with CLIP&lt;br&gt;
&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.techugo.com/blog/cost-to-develop-an-ecommerce-app-like-iconic-in-australia/" rel="noopener noreferrer"&gt;Techugo: cost to build an app like THE ICONIC in Australia&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://auspost.com.au/content/dam/auspost_corp/media/documents/ecommerce-reports/quarterly-ecommerce-reports/july-2026-quarterly-ecommerce-update.pdf" rel="noopener noreferrer"&gt;Australia Post quarterly ecommerce update (July 2026)&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://www.oaic.gov.au/privacy/australian-privacy-principles" rel="noopener noreferrer"&gt;OAIC: Australian Privacy Principles&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;Cost figures are indicative estimates from the sources above and vary by vendor, scope and region.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>machinelearning</category>
      <category>python</category>
      <category>ecommerce</category>
    </item>
  </channel>
</rss>
