<?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: Hashir</title>
    <description>The latest articles on DEV Community by Hashir (@offxhashir).</description>
    <link>https://dev.to/offxhashir</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%2F4150525%2F15d9c875-d6a1-4774-94a6-404befd15622.png</url>
      <title>DEV Community: Hashir</title>
      <link>https://dev.to/offxhashir</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/offxhashir"/>
    <language>en</language>
    <item>
      <title>Finding a hackathon idea took me 14 hours and 50 dead ideas. So I automated the killing</title>
      <dc:creator>Hashir</dc:creator>
      <pubDate>Sun, 25 Oct 2026 06:56:55 +0000</pubDate>
      <link>https://dev.to/offxhashir/finding-a-hackathon-idea-took-me-14-hours-and-50-dead-ideas-so-i-automated-the-killing-1e1i</link>
      <guid>https://dev.to/offxhashir/finding-a-hackathon-idea-took-me-14-hours-and-50-dead-ideas-so-i-automated-the-killing-1e1i</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-week3-2026-10-19"&gt;Hacktoberfest Open-Source AI Challenge Week 3: Slow and Steady&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;October 20. I needed one idea for this hackathon.&lt;/p&gt;

&lt;p&gt;Fourteen hours later I had killed roughly 50 of them by hand. Every single one already existed. I would think of something, search it four different ways, find the product that already did it, and start again.&lt;/p&gt;

&lt;p&gt;I asked three different AIs for help. Their five best pitches were all already built too. 0-for-5.&lt;/p&gt;

&lt;p&gt;The building was never the hard part. Finding something that wasn't already done was the whole day.&lt;/p&gt;

&lt;p&gt;So I built the thing that would have saved it.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Knell&lt;/strong&gt; — the bell that tolls for already-built ideas.&lt;/p&gt;

&lt;p&gt;You come in one of two ways:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Hackathon mode&lt;/strong&gt; — paste the challenge theme (plus your skills and how much time you have). Knell generates 8 candidates that fit it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Build an app mode&lt;/strong&gt; — you have no idea at all. You bring yourself: your stack, the worlds you know, your time, what you want out of it. Knell invents ideas at that crossing.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Then every candidate gets swept — three checks, live against the internet:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Existence&lt;/strong&gt; — is this already built? Searched the way a person would search it, plus targeted sweeps of product directories and app stores. The closest matches get their actual pages read, not just snippets.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Demand&lt;/strong&gt; — is anyone asking for this? Complaints, forum threads, one-star reviews of neighbouring products, "I wish there was an app that…" posts. Unbuilt can mean genius. It usually means nobody wants it.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Incumbent quality&lt;/strong&gt; — if it exists, is the existing version alive and good? A strong incumbent kills. A dead one is an opening.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Verdicts come back as DEAD, NO DEMAND, ZOMBIE GAP, or ALIVE. Survivors are ranked, with the evidence attached.&lt;/p&gt;

&lt;p&gt;The rule the whole product is built on: &lt;strong&gt;a verdict without a link is an opinion.&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;It's enforced in code, not in the prompt. A DEAD verdict only stands if it carries at least one real receipt from the actual search results, plus a cause of death. If the judge tries to kill an idea without evidence, the killer fields get wiped and the verdict doesn't render as a kill. The model is not allowed to invent your competition.&lt;/p&gt;

&lt;p&gt;Every kill goes into the &lt;strong&gt;graveyard&lt;/strong&gt;. Before any new check spends a single search, it looks in the graveyard first. Dead ideas stay dead — and they get cheaper every time.&lt;/p&gt;

&lt;p&gt;And before Knell had a name, I ran the sweep on names. It buried 9 of its own:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Verdikt&lt;/strong&gt; — killed by tryverdikt.app. A validator of the same name.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Unbuilt&lt;/strong&gt; — killed by unbuilt.me. "Don't build what already exists." Including this name.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;FlipKit&lt;/strong&gt; — killed by SnapSell. Photo to listing to posted, already done.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;ApplyOnce&lt;/strong&gt; — killed by a Chrome extension with the same name.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Morta&lt;/strong&gt; — killed by morta.io, a construction company.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Vetta&lt;/strong&gt; — killed by a recruitment AI.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mortem&lt;/strong&gt; — killed by the funeral-software crowd.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mortiq&lt;/strong&gt; — killed by MOTIQ, trademarked signage.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Premortem&lt;/strong&gt; — killed by being a method, not a name.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Knell survived. It means the bell rung for a death — which is the product: it tolls, and it shows you the body.&lt;/p&gt;

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

&lt;p&gt;Live on its .tech domain: &lt;a href="https://www.knellai.tech" rel="noopener noreferrer"&gt;https://www.knellai.tech&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Give it a theme and watch 8 candidates get generated — then watch them die one by one, each with the link that killed it. Or run Build an app mode with your own stack and see what survives.&lt;/p&gt;

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


&lt;div class="ltag-github-readme-tag"&gt;
  &lt;div class="readme-overview"&gt;
    &lt;h2&gt;
      &lt;img src="https://assets.dev.to/assets/github-logo-5a155e1f9a670af7944dd5e12375bc76ed542ea80224905ecaf878b9157cdefc.svg" alt="GitHub logo"&gt;
      &lt;a href="https://github.com/mateir0" rel="noopener noreferrer"&gt;
        mateir0
      &lt;/a&gt; / &lt;a href="https://github.com/mateir0/knell" rel="noopener noreferrer"&gt;
        knell
      &lt;/a&gt;
    &lt;/h2&gt;
    &lt;h3&gt;
      Knell is the death bell for already-built ideas. Give it a hackathon theme, or give it yourself, and it sweeps the internet for what already exists before naming what is worth building.
    &lt;/h3&gt;
  &lt;/div&gt;
  &lt;div class="ltag-github-body"&gt;
    
&lt;div id="readme" class="md"&gt;&lt;div class="markdown-heading"&gt;
&lt;h1 class="heading-element"&gt;Knell&lt;/h1&gt;
&lt;/div&gt;
&lt;p&gt;&lt;strong&gt;Knell is the death bell for already-built ideas.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;You give it either a hackathon theme, or yourself — a developer who can build
but has no idea what to build. Knell names candidate ideas, then sweeps the
internet for each one: does it already exist, is anyone asking for it, and is
the existing version any good. Ideas that are already built get killed, with
the receipt that killed them. Survivors come back ranked, with evidence.&lt;/p&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;The two modes&lt;/h2&gt;
&lt;/div&gt;
&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Mode&lt;/th&gt;
&lt;th&gt;You give it&lt;/th&gt;
&lt;th&gt;It gives you&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Hackathon&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;A theme (and optionally your skills, context and time)&lt;/td&gt;
&lt;td&gt;8 candidate ideas that fit the theme&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;&lt;strong&gt;Build an app&lt;/strong&gt;&lt;/td&gt;
&lt;td&gt;Your stack, what you know or care about, time, goal, platform&lt;/td&gt;
&lt;td&gt;8 candidate ideas that sit where your stack meets your worlds&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;
&lt;div class="markdown-heading"&gt;
&lt;h2 class="heading-element"&gt;Setup&lt;/h2&gt;
&lt;/div&gt;
&lt;p&gt;Requires Node 20+ and npm.&lt;/p&gt;
&lt;div class="highlight highlight-source-shell notranslate position-relative overflow-auto js-code-highlight"&gt;
&lt;pre&gt;npm install
cp .env.example .env.local   &lt;span class="pl-c"&gt;&lt;span class="pl-c"&gt;#&lt;/span&gt; optional: only if you want to change the model&lt;/span&gt;&lt;/pre&gt;…
&lt;/div&gt;&lt;/div&gt;
  &lt;/div&gt;
  &lt;div class="gh-btn-container"&gt;&lt;a class="gh-btn" href="https://github.com/mateir0/knell" rel="noopener noreferrer"&gt;View on GitHub&lt;/a&gt;&lt;/div&gt;
&lt;/div&gt;


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

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Next.js 15 + TypeScript + Tailwind v4&lt;/strong&gt; for the app.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Mastra&lt;/strong&gt; runs the pipeline as two typed workflows: &lt;code&gt;knell-generate&lt;/code&gt;, and &lt;code&gt;knell-check&lt;/code&gt; (write queries → build evidence → judge → validate). Failures travel through the workflow as typed data and get restored by the runner, instead of being swallowed as empty results.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;SerpApi&lt;/strong&gt; does the sweep, with Serper as fallback — the check has to see what a human Googling would see, because that manual Google is exactly the chore being automated. Search budget is enforced in code (≤28 searches per run), because search quota is the product's oxygen.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The model layer is just an endpoint.&lt;/strong&gt; Generation and judging are separate calls to any OpenAI-compatible endpoint, set by environment variables. The repo defaults to a local open-weight model — Gemma running in Ollama, $0, offline. The live site runs a hosted model through the exact same switch. No code cares which one answers.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;MongoDB Atlas&lt;/strong&gt; holds the graveyard and run history — the agent's long-term memory. Cache-before-search lives here.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Sentry agent tracing&lt;/strong&gt; covers every run end to end — generate, search, fetch, judge, graveyard spans — so any verdict can be audited afterwards, step by step, instead of trusted.&lt;/li&gt;
&lt;/ul&gt;

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

&lt;p&gt;Because a closed API would make this product a hostage.&lt;/p&gt;

&lt;p&gt;Knell's judge has one job: kill ideas honestly. If that judge is a single vendor's model behind a single vendor's API, then my product's honesty is whatever their pricing, their rate limits, and their model's mood say it is this month. I can't swap it, I can't run it offline, and I definitely can't fine-tune it on my own kill data.&lt;/p&gt;

&lt;p&gt;Open weights and an open harness fix all three:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Swap, don't rewrite.&lt;/strong&gt; The model is an environment variable. Local Gemma for free and private, a hosted model when I want speed — same code, same prompts, same verdict format.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;It can run on a laptop with no internet.&lt;/strong&gt; The default setup in the repo is Ollama + Gemma. Your half-formed ideas never have to leave your machine to get judged.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The judge can become mine.&lt;/strong&gt; Every real verdict Knell issues is a labeled training example. The repo ships the full loop — export the graveyard as training data, fine-tune a LoRA judge on Tinker, evaluate it against the current judge on verdict agreement and receipt validity — and the rule is written down: the tuned judge only ships if it beats the baseline on both. That loop only exists because the model layer is open. You cannot fine-tune a closed API on 50 corpses and keep the result.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Honest part: the fine-tune itself hasn't run yet. The pipeline is built and the data is accumulating; the run happens when the dataset is worth it, and the numbers will say whether it ships. I'm not claiming a trained judge I don't have.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Lazy Payoff
&lt;/h2&gt;

&lt;p&gt;The numbers, all measured on the live build:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;Finding one idea by hand: &lt;strong&gt;14 hours, ~50 manual kills.&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;One Knell run: &lt;strong&gt;8 candidates generated and swept in minutes.&lt;/strong&gt;
&lt;/li&gt;
&lt;li&gt;A fresh check on one idea: &lt;strong&gt;10.1 seconds&lt;/strong&gt; to judge and kill.&lt;/li&gt;
&lt;li&gt;The same idea checked again: &lt;strong&gt;1.2 seconds&lt;/strong&gt; — the graveyard answered from memory and spent zero searches.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;And yes, I used it this week — on itself. The name massacre above is a real Knell output, and the graveyard on the live site is the real kill list, not demo data.&lt;/p&gt;

&lt;p&gt;That's the sloth philosophy applied to the worst chore in building: don't spend your energy finding out the hard way. Let the bell ring first.&lt;/p&gt;

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

&lt;ul&gt;
&lt;li&gt;Best Use of Mastra&lt;/li&gt;
&lt;li&gt;Best Use of MongoDB Atlas&lt;/li&gt;
&lt;li&gt;Best Use of SerpApi&lt;/li&gt;
&lt;li&gt;Best Use of Sentry Agent Tracing&lt;/li&gt;
&lt;li&gt;Best Use of a .tech Domain&lt;/li&gt;
&lt;/ul&gt;

</description>
      <category>devchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>I built my best friend a mistake engine for the NUST entry test</title>
      <dc:creator>Hashir</dc:creator>
      <pubDate>Sun, 04 Oct 2026 08:53:26 +0000</pubDate>
      <link>https://dev.to/offxhashir/i-built-my-best-friend-a-mistake-engine-for-the-nust-entry-test-434d</link>
      <guid>https://dev.to/offxhashir/i-built-my-best-friend-a-mistake-engine-for-the-nust-entry-test-434d</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hacktoberfest-weekend-2026-10-01"&gt;Hacktoberfest Weekend Challenge: Build for a Friend&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;My best friend SMS is studying for the NUST entry test — the exam that decides whether he gets into Pakistan's top engineering university. For months he cycled through quiz apps, and they all had the same flaw: they'd generate questions, he'd answer them, and the app would forget everything by the next session. Nobody remembered what he got wrong. So he kept losing the same marks twice.&lt;/p&gt;

&lt;p&gt;I built &lt;strong&gt;NEVERTWICE&lt;/strong&gt; to fix exactly that. It's a study app with a memory. The loop is simple:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;mistakes → weakness graph → targeted drill → graph update&lt;/strong&gt;&lt;/p&gt;

&lt;p&gt;You answer questions in a drill. Get one wrong and you classify it by topic with one tap. The weakness graph updates — your threats, ranked worst first. The next drill pulls hardest from the top of that graph. Nail the topic next time and the threat shrinks. Four tabs, zero clutter: DRILL | THREATS | PROGRESS | FILES. Built phone-first, because that's where SMS studies.&lt;/p&gt;

&lt;p&gt;There's also a full 200-question, 180-minute mock for exam-day simulation, and a trajectory view so he can watch his accuracy climb across sessions instead of guessing whether he's improving.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Saturday test
&lt;/h2&gt;

&lt;p&gt;I didn't build this in a vacuum. On Saturday, October 3rd, I handed the app to SMS with one rule: if the drills don't feel uncannily personal — if it can't show him things he actually got wrong — we kill the concept.&lt;/p&gt;

&lt;p&gt;It passed. And he broke three things in the process, all fixed the same day:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Classifying mistakes was too slow.&lt;/strong&gt; Tapping through categories after every wrong answer killed his flow. Now it's one tap, with a 700ms auto-advance.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Explanations referenced answer letters.&lt;/strong&gt; "The correct answer is B" means nothing when the options shuffle. Now the debrief names the correct answer by its content, not its letter.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Motion sensitivity.&lt;/strong&gt; A full motion pass plus &lt;code&gt;prefers-reduced-motion&lt;/code&gt; support, because a study app shouldn't make anyone dizzy.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Here's what he said after the session:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;"I definitely see myself using your app to prepare for my future NUST entry test cause on this app i can track my progress via a graph and i can also solve questions from the real NUST NET type mock exam allowing me to understand the depthness of exam questions before I actually attempt the real test"&lt;/p&gt;

&lt;p&gt;— SMS, my best friend and the reason this exists&lt;/p&gt;
&lt;/blockquote&gt;

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

&lt;p&gt;  &lt;iframe src="https://www.youtube.com/embed/4tvDKO-U0Sg" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;p&gt;Try it live — no signup, no paywall, no ads: &lt;strong&gt;&lt;a href="https://nevertwice.vercel.app" rel="noopener noreferrer"&gt;https://nevertwice.vercel.app&lt;/a&gt;&lt;/strong&gt;&lt;/p&gt;

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

&lt;p&gt;&lt;strong&gt;&lt;a href="https://github.com/mateir0/nevertwice" rel="noopener noreferrer"&gt;https://github.com/mateir0/nevertwice&lt;/a&gt;&lt;/strong&gt; — public repo, MIT licensed.&lt;/p&gt;

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

&lt;p&gt;Next.js on Vercel, with question generation running server-side through Groq on an open-weight model (&lt;code&gt;openai/gpt-oss-120b&lt;/code&gt;). The API key never touches the client.&lt;/p&gt;

&lt;p&gt;A few things I'm proud of, stated plainly:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;It works with no internet.&lt;/strong&gt; A 120-question verified bank ships with the app, and a service worker drops it into offline bank mode. Drills run 100% on the bank with zero AI when you're offline.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Your data is yours.&lt;/strong&gt; Mistake history and the weakness graph live in browser storage. Dossier export/import means you back up or move everything yourself.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Honest fallbacks.&lt;/strong&gt; When AI generation fails or quotas run out, the app says so — "OFFLINE — BANK MODE", "DAILY PRINT QUOTA SPENT — DRILLS UNAFFECTED" — instead of silently degrading into repeats.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Security taken seriously.&lt;/strong&gt; Prompt-injection guards on the generation taxonomy, prototype-pollution rejection, body-size limits, security headers, server-side rate limits. There's a SECURITY.md in the repo.&lt;/li&gt;
&lt;li&gt;&lt;strong&gt;74/74 tests green.&lt;/strong&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;One honesty note, because it matters: NUST doesn't release official past papers, and the app says so in its own UI. The mock is modeled on the NET format. It doesn't pretend to be leaked papers.&lt;/p&gt;

&lt;p&gt;Built solo over a weekend with AI pair-programming — the AI handled the boilerplate, I handled the judgment calls. The best feature wasn't anything I planned: watching a red threat shrink because you finally nailed a topic three sessions in a row. I only discovered that by watching SMS use it.&lt;/p&gt;

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

&lt;p&gt;The challenge asks this directly, so here's the honest answer for NEVERTWICE:&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does it run with no internet?&lt;/strong&gt; Yes — the offline bank mode exists because of this. The 120-question bank and the service worker mean SMS can drill on a bus with no signal. A closed API-only app couldn't do that.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does it keep his data off servers he doesn't control?&lt;/strong&gt; His mistake history — the most personal data in the app, a map of everything he doesn't know — never leaves his browser. Only the generation prompts go to Groq's cloud. If he doesn't trust that, the bank works fully offline.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Can the model be swapped?&lt;/strong&gt; Yes, and that's the point of building on an open-weight model. The question validator and the fallback logic don't care which model generates the questions. If Groq throttles, the bank carries him. If a better open model appears tomorrow, it's a config change, not a rewrite. No vendor lock-in on the intelligence.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Does it cost him anything?&lt;/strong&gt; Nothing. Free tier for generation, free bank offline, free hosting. A gift for a friend shouldn't come with a subscription.&lt;/p&gt;

&lt;p&gt;Open didn't just make this cheaper to build. It made it &lt;em&gt;his&lt;/em&gt; — inspectable, forkable, exportable, and usable with zero infrastructure. For a study app holding someone's exam future, that ownership isn't a nice-to-have. It's the whole point.&lt;/p&gt;

&lt;h2&gt;
  
  
  My Agent Session
&lt;/h2&gt;

&lt;p&gt;Built with OpenCode and Muse Spark 1.3, one consolidated prompt per task. The full session record — real transcripts, verbatim quotes, no reconstruction — is in the repo: &lt;a href="https://github.com/mateir0/nevertwice/blob/master/agent-session.md" rel="noopener noreferrer"&gt;agent-session.md&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Five moments tell the story: the drill engine build, the SMS feedback fixes, the mock Groq-budget fix, the security audit, and the structured logging. The drill planner learned to write dossier-voiced reasons like &lt;em&gt;"Quadratic Equations keeps bleeding — 6 misreads in 3 days. 7 questions engineered to punish skimming."&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;My favorite exchange: during the mock-fix verification, a run hit a 19-minute Groq throttle, and the call was &lt;em&gt;"STOP the live mock verification (kill process 2992) — do NOT chase the 'no 429' criterion."&lt;/em&gt; Knowing when to stop verifying is engineering too.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;For SMS: never twice.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;AI disclosure: built with OpenCode and Muse Spark 1.3; AI assistance was used to prepare this submission.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>weekendchallenge</category>
      <category>hf26challenge</category>
    </item>
    <item>
      <title>GrantQuest — I Turned Scholarship Hunting into an RPG</title>
      <dc:creator>Hashir</dc:creator>
      <pubDate>Thu, 01 Oct 2026 16:09:37 +0000</pubDate>
      <link>https://dev.to/offxhashir/grantquest-i-turned-scholarship-hunting-into-an-rpg-3b4i</link>
      <guid>https://dev.to/offxhashir/grantquest-i-turned-scholarship-hunting-into-an-rpg-3b4i</guid>
      <description>&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%2F2j0zrqlxg7vxts7tzaic.jpg" 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%2F2j0zrqlxg7vxts7tzaic.jpg" alt="GrantQuest — I turned scholarship hunting into an RPG" width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;This is a submission for the Sanity Challenge, Path Two: Vibe-Code Something Strange&lt;/em&gt;&lt;/p&gt;

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

&lt;p&gt;GrantQuest — scholarship hunting, but it's an RPG quest log.&lt;/p&gt;

&lt;p&gt;Every scholarship is a quest. Eligibility requirements are gates you clear one by one. Documents are inventory you gather. Deadlines are countdown timers that do not care about your feelings. Clearing gates earns XP, and your Quest Log tracks every application like a campaign journal.&lt;/p&gt;

&lt;p&gt;But here's the twist that makes it more than a theme: scholarship data rots. Deadlines change, programs get discontinued, amounts get updated every year — and half the scholarship sites on the internet are quietly serving you last year's facts. So every quest in GrantQuest carries a freshness badge: FRESH (verified within 30 days and the deadline still ahead) or NEEDS RE-VERIFICATION. The data's trustworthiness is modeled as data, not vibes.&lt;/p&gt;

&lt;p&gt;25 real scholarships. Real deadlines. Real money. No login, no signup — just open it and start your expedition.&lt;/p&gt;

&lt;p&gt;It's for students like me who are drowning in 40 open tabs trying to figure out if they're eligible for anything.&lt;/p&gt;




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

&lt;p&gt;&lt;strong&gt;Play it here:&lt;/strong&gt; &lt;a href="https://grantquest.tech" rel="noopener noreferrer"&gt;https://grantquest.tech&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Video walkthrough (60 seconds):&lt;/strong&gt;&lt;br&gt;
  &lt;iframe src="https://www.youtube.com/embed/3JO1z3NQTP8" width="710" height="399"&gt;
  &lt;/iframe&gt;
&lt;/p&gt;

&lt;p&gt;There's also a 30-second guided tour built right into the site — hit "Take the 30-second tour" on the hero and it walks you through the whole loop.&lt;/p&gt;




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

&lt;p&gt;&lt;a href="https://github.com/mateir0/grandquest" rel="noopener noreferrer"&gt;https://github.com/mateir0/grandquest&lt;/a&gt;&lt;/p&gt;




&lt;h2&gt;
  
  
  My Build Process
&lt;/h2&gt;

&lt;p&gt;I built this with OpenCode running Muse Spark, one prompt at a time, fifteen prompts total. The plan was simple: quest board, quest detail pages, an eligibility engine, a quest log with XP, and a freshness-verification workflow in the Studio. Then the stretch goals: derived freshness badges, the official Workflows API, a deadline-sweeper Function, an App SDK board, and a guided tour for judges.&lt;/p&gt;

&lt;p&gt;Here's where it went wrong, because that's the interesting part:&lt;/p&gt;

&lt;h3&gt;
  
  
  The deploy that 404'd everything
&lt;/h3&gt;

&lt;p&gt;My Next.js app imported shared logic from a folder above web/. Worked perfectly on my machine. Vercel 404'd the entire site, because Vercel only uploads the web/ directory and my import pointed at files that didn't exist up there. I vendored the file into web/lib/, redeployed, done.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Lesson now burned into my brain: never import above web/ in a Next.js app on Vercel.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h3&gt;
  
  
  The freshness rule I overcomplicated
&lt;/h3&gt;

&lt;p&gt;My first version said a quest goes stale if its deadline is within 60 days. Then I stared at it and realized: that's the countdown's job, not the freshness badge's job. Freshness answers "is this data still true?" Urgency answers "is the deadline close?" Two different questions, two different UI elements. Simplified the rule to: verified within 30 days AND deadline not passed = FRESH. Everything else gets flagged. The code got shorter and the concept got sharper, which is usually a sign you're done.&lt;/p&gt;

&lt;h3&gt;
  
  
  The ERSU incident — my favorite failure
&lt;/h3&gt;

&lt;p&gt;I had a test record for an Italian scholarship (ERSU Messina) sitting in the dataset with half-guessed data. The lazy move would've been stamping it verified and moving on — who's going to check, right? Instead I went and read the actual official bando, in Italian, from ersumessina.it, and replaced every guessed field with real numbers: real amounts (€7,171.11 for fuori sede), real ISEE thresholds, real dates. And it now correctly shows NEEDS RE-VERIFICATION, because applications closed back in August.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;An honest red badge beats a lying green one. That record is the whole philosophy of the app in one card.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h3&gt;
  
  
  The Vanier plot twist
&lt;/h3&gt;

&lt;p&gt;While researching 10 new scholarships to add before submission, I discovered that the Vanier Canada Graduate Scholarship — one of the most famous scholarships on the planet — is officially discontinued. "No longer accepting applications," right there on the government website. So I replaced it with its actual successor, the Canada Graduate Research Scholarship–Doctoral. The freshness system exists because of exactly this kind of thing: the ground moves under you and nobody tells you.&lt;/p&gt;

&lt;h3&gt;
  
  
  The janky tour
&lt;/h3&gt;

&lt;p&gt;First version of the guided judge tour felt laggy when I watched it back. You know the feeling — something's just slightly off and you can't unsee it. Rebuilt the animation approach until it was smooth.&lt;/p&gt;

&lt;h3&gt;
  
  
  The leaked token
&lt;/h3&gt;

&lt;p&gt;My final security sweep found a real Sanity write token sitting in a committed .env.example. Sanitized it to a placeholder, revoked the token, rotated a fresh one. (The old one is still in git history, but it's dead, so it doesn't matter.)&lt;/p&gt;




&lt;h2&gt;
  
  
  Agent Sessions
&lt;/h2&gt;

&lt;p&gt;Fifteen prompts, every wrong turn included — the unedited build transcripts, exactly as they happened:&lt;/p&gt;

&lt;p&gt;&lt;a href="https://github.com/mateir0/grandquest/blob/main/agent-session-embed.md" rel="noopener noreferrer"&gt;Read the full build sessions on GitHub&lt;/a&gt;&lt;/p&gt;




&lt;h2&gt;
  
  
  Where I reached past the Studio
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Workflows (official API):&lt;/strong&gt; every quest moves unverified → underReview → verified through the real Workflows engine — not a status dropdown, actual workflow transitions. All 25 quests were driven through real transitions.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Functions:&lt;/strong&gt; a scheduled Function sweeps daily. Any quest whose deadline passed, or whose verification is older than 30 days, gets flagged automatically — with an immutable audit log of every sweep.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;App SDK:&lt;/strong&gt; the Quest Master's Board, a dashboard app for working through the verification queue.&lt;/p&gt;




&lt;h2&gt;
  
  
  Sanity Project Details
&lt;/h2&gt;

&lt;p&gt;Project ID: &lt;code&gt;aitdwcxh&lt;/code&gt; · Dataset: &lt;code&gt;production&lt;/code&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Public dataset — query it yourself:&lt;/strong&gt;&lt;br&gt;
&lt;a href="https://aitdwcxh.api.sanity.io/v2024-01-01/data/query/production?query=*%5B_type%3D%3D%22quest%22%5D" rel="noopener noreferrer"&gt;https://aitdwcxh.api.sanity.io/v2024-01-01/data/query/production?query=*%5B_type%3D%3D%22quest%22%5D&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Schema highlights: quests with structured eligibility gates (not prose — actual typed data with requirements), documents as inventory items, deadlines, award amounts, and the verification workflow state machine. Gates being data instead of text is the whole trick — it's what lets the app compute your eligibility instead of just displaying a wall of text.&lt;/p&gt;

</description>
      <category>devchallenge</category>
      <category>sanitychallenge</category>
      <category>sanity</category>
      <category>ai</category>
    </item>
  </channel>
</rss>
