<?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: Quang Vinh Le</title>
    <description>The latest articles on DEV Community by Quang Vinh Le (@quangvinhlee).</description>
    <link>https://dev.to/quangvinhlee</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%2F4158569%2Ff493eef5-a489-4daf-b555-8bf7d954103a.png</url>
      <title>DEV Community: Quang Vinh Le</title>
      <link>https://dev.to/quangvinhlee</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/quangvinhlee"/>
    <language>en</language>
    <item>
      <title>FlipCheck: Helping an Electronics Repairer Find Deals with Local AI</title>
      <dc:creator>Quang Vinh Le</dc:creator>
      <pubDate>Sun, 18 Oct 2026 12:48:31 +0000</pubDate>
      <link>https://dev.to/quangvinhlee/flipcheck-helping-an-electronics-repairer-find-deals-with-local-ai-23o6</link>
      <guid>https://dev.to/quangvinhlee/flipcheck-helping-an-electronics-repairer-find-deals-with-local-ai-23o6</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12"&gt;Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;A friend of mine repairs electronics and resells them. He regularly checks Facebook Marketplace, eBay, and other platforms for used devices, repairable products, and spare parts. A good purchase might be something he can resell, something he can fix, or a donor device with parts for another repair.&lt;/p&gt;

&lt;p&gt;The searching takes time: repeat the same queries, open listings, check the photos, work out the specifications, and decide which sellers are worth contacting.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;I built FlipCheck to bring those searches and listing details into one workspace.&lt;/strong&gt; It uses Gemma 4 locally to extract product information from seller descriptions, with supporting text beside each assessment.&lt;/p&gt;

&lt;p&gt;The current version connects &lt;strong&gt;eBay Australia and nearby Facebook Marketplace&lt;/strong&gt;. You can save searches for a pc, a MacBook, and a camera, then scan them together. Each search has its own budget, target margin, and cost assumptions. Facebook searches use a chosen location and pickup radius.&lt;/p&gt;

&lt;p&gt;From the combined feed, you can:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Browse actual seller photos and open the original listings.&lt;/li&gt;
&lt;li&gt;Review extracted specifications, supporting excerpts, and missing details.&lt;/li&gt;
&lt;li&gt;Correct facts, save or dismiss listings, and inspect margin estimates when enough evidence is available.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The potential paying customer is someone running a small electronics repair and resale business. The value I want to test is whether this reduces repetitive checking and helps them decide which listings deserve a closer look.&lt;/p&gt;

&lt;p&gt;This is a working local MVP. Scans are started manually. Faulty and parts-only products are currently excluded from normal resale valuation; a dedicated repair and parts workflow is a planned extension inspired by my friend's work.&lt;/p&gt;

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

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/fQGwvDcWVOQ" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;p&gt;The walkthrough shows real marketplace results, nearby search controls, seller photos, and the evidence behind the assessments. It includes synthetic narration and captions.&lt;/p&gt;

&lt;p&gt;FlipCheck runs locally. To try it, follow the &lt;a href="https://github.com/quangvinhlee/FlipCheck#readme" rel="noopener noreferrer"&gt;repository setup instructions&lt;/a&gt;. Live discovery requires a SerpApi key for eBay and an Apify token for Facebook.&lt;/p&gt;

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

&lt;p&gt;&lt;a href="https://github.com/quangvinhlee/FlipCheck" rel="noopener noreferrer"&gt;&lt;strong&gt;View FlipCheck on GitHub&lt;/strong&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The application source is MIT-licensed. Model weights, provider services, and marketplace data retain their respective terms.&lt;/p&gt;

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

&lt;p&gt;&lt;strong&gt;Gemma 4 runs locally through Ollama&lt;/strong&gt;, using &lt;code&gt;gemma4:e2b&lt;/code&gt;. Next.js, React, and TypeScript provide the interface. A Python backend handles marketplace discovery, validation, cost calculations, and local SQLite storage.&lt;/p&gt;

&lt;p&gt;SerpApi supplies eBay results, and Apify supplies Facebook Marketplace results. The backend normalises the listings and retains the seller photos. Gemma then extracts the product details from the listing text.&lt;/p&gt;

&lt;p&gt;Each extracted fact must include a supporting source excerpt. Python checks those excerpts, required specifications, and conflicting details before valuation. Product categories have different requirements: a camera body needs different comparisons from a kit with a lens, while a gaming laptop needs its GPU configuration.&lt;/p&gt;

&lt;p&gt;For nearby Facebook searches, location access happens only after a button click. The app uses rounded coordinates and approximate straight-line distances to filter listings within the selected radius.&lt;/p&gt;

&lt;p&gt;One useful lesson came from seeing almost everything labelled &lt;strong&gt;“Needs review.”&lt;/strong&gt; That label covered several problems: unknown condition, missing specifications, unknown shipping or pickup costs, and missing completed-sale evidence. I made those reasons visible so the user knows what to check next.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Live searches collect asking prices. Verified completed-sale comparisons must currently be imported.&lt;/strong&gt; The “Worth a closer look” badge requires at least five recent matching verified sales, confirmed product details, known collection/shipping costs, and a margin that meets the target using the lower-quartile resale estimate. A user-entered resale assumption is labelled separately and does not qualify for that badge.&lt;/p&gt;

&lt;p&gt;Validation included 59 passing Python tests, TypeScript checks, a production build, and browser checks covering desktop/mobile layouts, accessibility, location controls, and photo galleries. Both marketplace connectors have been exercised with real data; resale accuracy and realised profit still need real-world validation.&lt;/p&gt;

&lt;p&gt;I used an AI coding assistant during development, debugging, test writing, and preparation of this submission and demo.&lt;/p&gt;

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

&lt;p&gt;&lt;strong&gt;Repeated analysis does not require a hosted model charge for every listing.&lt;/strong&gt; I can adjust a prompt or product profile and run extraction again locally. That matters for a tool intended to screen listings regularly, although hardware, electricity, and discovery services still have costs.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;I can adapt the model's behaviour to the job.&lt;/strong&gt; The model choice, prompts, extraction schema, and validation rules are under my control. That gives me room to develop a repair and parts workflow around how my friend evaluates purchases.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Listing analysis and records can stay on the user's machine.&lt;/strong&gt; With the model downloaded, imported listing text can be processed offline. Corrections and decisions are stored in local SQLite. Live discovery still needs internet: search terms and rounded search coordinates go to the relevant providers, place lookup uses Nominatim, and photos load from marketplace servers.&lt;/p&gt;

&lt;p&gt;For FlipCheck, the benefit of open weights is control over where inference runs, how often it runs, and how it can be adapted. I have not fine-tuned the model or measured it against a closed alternative.&lt;/p&gt;

&lt;h2&gt;
  
  
  My First User
&lt;/h2&gt;

&lt;p&gt;I showed FlipCheck to the friend whose repair and resale work inspired it. He thought the app was already good and said he would pay for a more complete version with features better suited to his job.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;He has not paid yet.&lt;/strong&gt; This is early feedback and conditional willingness to pay. I have not measured time savings or completed a structured sourcing trial with him.&lt;/p&gt;

&lt;p&gt;His feedback gives me a direction: develop the repair and parts workflow, agree on the features that matter most to him, and have him try it during a real sourcing session. The next question is whether he keeps using it and finds enough value to pay.&lt;/p&gt;

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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Gemma:&lt;/strong&gt; Local product extraction with supporting seller text is central to the workflow.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of SerpApi:&lt;/strong&gt; Live eBay Australia discovery brings real listings into the saved product searches.&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
      <category>ai</category>
      <category>opensource</category>
    </item>
    <item>
      <title>Outwards Days — make a little room for outside</title>
      <dc:creator>Quang Vinh Le</dc:creator>
      <pubDate>Sun, 11 Oct 2026 12:25:13 +0000</pubDate>
      <link>https://dev.to/quangvinhlee/outwards-days-make-a-little-room-for-outside-l1n</link>
      <guid>https://dev.to/quangvinhlee/outwards-days-make-a-little-room-for-outside-l1n</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week1-2026-10-05"&gt;Hacktoberfest Open-Source AI Challenge Week 1: Touch Grass&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;Outwards Days helps someone with a crowded week find a realistic reason to go outside. It starts with their actual schedule: work, busy blocks, lunch gaps, preferred times, travel limits, and the days they really have free. It can suggest a short walk or a little garden care on a workday, then make room for a longer outing when the whole day allows it. The goal is to make the screen the shortest part of the experience.&lt;/p&gt;

&lt;p&gt;Each suggestion has a time budget that includes preparation and travel, a category, and a concrete activity. When place data is available, it can name a nearby destination instead of saying only “go to a park.” Small missions can ask for an observation, a written reflection, a simple action, or a photo. For a photo mission, the person can choose another visible subject if the original is not on the route. Completed places, categories, missions, and notes help vary future plans; a return to the garden might ask how an earlier planting has grown.&lt;/p&gt;

&lt;p&gt;People can review a proposed week, accept or move an outing, and record what actually happened. Guest profiles can be upgraded to an email/password account so their server-side schedule, plans, and history follow that account.&lt;/p&gt;

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

&lt;p&gt;[Watch or download the narrated demo (MP4)]  &lt;iframe src="https://www.youtube.com/embed/1qa1oF600s0" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;p&gt;The 40-second video shows the Android emulator: preferences, schedule review, the generated week, outing budgets and missions, nearby places, and progress. It uses the local backend and model. It does &lt;strong&gt;not&lt;/strong&gt; show a hosted browser deployment, a live model request being generated on camera, or a real outdoor trip. The browser release has been prepared and tested locally, but it has not yet been verified at a public URL. A physical-phone camera-to-verdict test and a real outdoor trial are still outstanding.&lt;/p&gt;

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

&lt;p&gt;&lt;a href="https://github.com/quangvinhlee/Outwards-Days" rel="noopener noreferrer"&gt;Source code and local setup&lt;/a&gt; · &lt;a href="https://github.com/quangvinhlee/Outwards-Days/blob/main/docs/architecture.md" rel="noopener noreferrer"&gt;Architecture&lt;/a&gt; · &lt;a href="https://github.com/quangvinhlee/Outwards-Days/blob/main/evals/results.json" rel="noopener noreferrer"&gt;Recorded evaluation&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;The repository includes the Expo app, Fastify API, deterministic scheduler, Mastra workflow, tests, evaluation data, and deployment guide. The intended public release is the browser app; the narrated walkthrough was captured from the Android build of the shared interface.&lt;/p&gt;

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

&lt;p&gt;The interface uses Expo and React Native. Fastify handles sessions, plans, acceptance, completion, and owner-scoped MongoDB data. A Mastra workflow gathers schedule, weather, place, and bounded history context, then asks &lt;strong&gt;Gemma 4 E2B&lt;/strong&gt; through Ollama to choose and personalize eligible outdoor activities.&lt;/p&gt;

&lt;p&gt;The model does not get to decide whether a suggestion fits. Code calculates the free intervals, travel allowance, duration, daylight, weather limits, and activity frequency. Model output is parsed and checked against those rules. If it fails, the workflow makes one repair attempt; if it still fails, the app labels its basic catalogue suggestions as a fallback. Acceptance checks the schedule again, since it may have changed after generation. This keeps an interesting suggestion from becoming a conflicting calendar commitment.&lt;/p&gt;

&lt;p&gt;I recorded a small, five-case local evaluation with &lt;code&gt;gemma4:e2b&lt;/code&gt;. Three cases attempted inference: two produced validated Gemma plans in 2.390 and 13.429 seconds, and one used the labelled basic fallback after two rejected attempts. Two infeasible cases correctly returned no activities without inference. This is functional evidence for those cases, not a reliability benchmark or proof of hosted performance.&lt;/p&gt;

&lt;p&gt;The app can retain plans and queue completions locally. Stable event IDs make a repeated sync request idempotent on the server. &lt;a href="https://dev.to/salazarismo/the-hidden-problems-of-offline-first-sync-idempotency-retry-storms-and-dead-letters-1no8"&gt;Salazar's discussion of offline retry failures&lt;/a&gt; reinforced the value of keeping failed work visible; &lt;a href="https://dev.to/rcortez056/why-probabilistic-ai-needs-deterministic-control-g8d"&gt;this discussion of deterministic AI control&lt;/a&gt; echoes the boundary between model suggestions and code-enforced rules. These are design references, not claims that Outwards Days implements their entire architectures.&lt;/p&gt;

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

&lt;p&gt;Gemma is an open-weight model I can run through Ollama, inspect, and replace without outsourcing the planning decision to a closed API. Mastra and the scheduling code make the workflow and its limits visible. That separation matters here: the model can make an outing feel personal, while the program remains responsible for whether there is actually time to do it. The application code is MIT-licensed; Gemma has its own model terms.&lt;/p&gt;

&lt;p&gt;This is not an offline AI app on the phone. Generating a new plan needs the backend and inference service, and weather and place discovery need network services. The browser stores a cached plan and completion queue while loaded, but it has no service worker to guarantee reopening offline. Native local reminders are not part of the browser release.&lt;/p&gt;

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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Gemma:&lt;/strong&gt; Gemma 4 E2B supplies the activity choices and personalized wording. The repository includes the real Ollama integration, model digest, and local evaluation results.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of Mastra:&lt;/strong&gt; The planning workflow coordinates context gathering, model generation, validation, bounded repair, and a clearly labelled fallback.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Best Use of GitHub Copilot (GitHub Actions route):&lt;/strong&gt; &lt;a href="https://github.com/quangvinhlee/Outwards-Days/actions/runs/38137336342" rel="noopener noreferrer"&gt;GitHub Actions&lt;/a&gt; automatically checks types, tests, builds, dependency compatibility, and Expo native exports on pushes and pull requests. I used Actions automation for this category; I am not claiming to have built the app with Copilot.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Render, DigitalOcean, and MongoDB Atlas are planned for the hosted release, so I am not entering their categories on deployment plans alone. AI assistance was used in building the project and drafting this post.&lt;/p&gt;

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