<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Building Nyata]]></title><description><![CDATA[How Nyata is built: a hyperlocal AI companion for free meals, local support and services. Notes on LLMs, APIs, MCP and accessibility, city by city.]]></description><link>https://nyata.hashnode.dev</link><image><url>https://cdn.hashnode.com/uploads/logos/6ae66e5be4d1e1fdccae2631/b85227d8-cfdd-4456-bf04-75c849e64308.png</url><title>Building Nyata</title><link>https://nyata.hashnode.dev</link></image><generator>RSS for Node</generator><lastBuildDate>Sun, 22 Nov 2026 17:14:26 GMT</lastBuildDate><atom:link href="https://nyata.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Why I'm building Nyata, and why I'm writing it down]]></title><description><![CDATA[A hyperlocal AI companion that turns "I need help" into the right place nearby, and the engineering notes behind it.]]></description><link>https://nyata.hashnode.dev/why-i-m-building-nyata-and-why-i-m-writing-it-down</link><guid isPermaLink="true">https://nyata.hashnode.dev/why-i-m-building-nyata-and-why-i-m-writing-it-down</guid><dc:creator><![CDATA[Building Nyata]]></dc:creator><pubDate>Wed, 04 Nov 2026 09:35:08 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6ae66e5be4d1e1fdccae2631/179e6f60-e889-462b-899b-3acedd35ab5c.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>"Where can I get a hot meal tonight?"</p>
<p>It's one of the simplest questions a person can ask, and one of the hardest to get a straight answer to. The information exists, but it's scattered across leaflets, charity websites, Facebook posts and word of mouth. Half of it is out of date, and almost none of it tells you what's open <em>right now</em>, <em>near you</em>.</p>
<p>Nyata is my attempt to fix that. This blog is where I'll write down how it's built.</p>
<h2>Where it started</h2>
<p>Before Nyata, I worked in frontline welfare support with vulnerable adults. A lot of that job wasn't the support itself. It was finding out where the support was: which meal service runs on a Tuesday, where someone can get a shower, which group meets this week and whether they still take walk-ins.</p>
<p>People who most needed that information had the least time, battery and patience to dig for it. And the people helping them, me included, were often working from a printout that had changed since it was printed.</p>
<p>The idea became my dissertation at UWE Bristol, on AI in business. The dissertation became a company, Nyata Ltd. The company became an app that's now live on iOS, Android and Mac.</p>
<h2>What Nyata does</h2>
<p>You tell Nyata your situation in your own words, typed or spoken. It either gives you useful advice, or points you to the organisation nearby that can actually help, with the day, the time and how to get there.</p>
<p>That covers the things I saw people struggle to find:</p>
<ul>
<li>free meals, showers, clothing and shelter</li>
<li>support groups and employment support</li>
<li>charities and community services, which we verify before they appear</li>
</ul>
<p>And the everyday local things everyone needs: barbers, beauty, cleaning, local events, and taxi booking, which we're currently testing. There's a community chat too (with mini-games, because people stay where they have fun), and in-chat reminders that nudge you 30 minutes before an appointment.</p>
<p>There's no premium tier. The core of it is free.</p>
<h2>How it's built, at a glance</h2>
<p>The longer write-ups will come post by post. Here's the shape of it.</p>
<p><strong>One backend, every platform.</strong> The iOS, Android and Mac apps all run on the same PHP and MySQL backend. One fix ships everywhere at once, which matters a lot when you're a small team.</p>
<p><strong>More than one model.</strong> The intelligence layer is deliberately multi-model. An LLM fine-tuned on local-service intent handles "where can I find…" questions, and other models drive features where a different strength matters. Every character is told to answer briefly and exactly what was asked, then offer to go deeper. Short answers are kinder on a cracked phone screen and on a token budget.</p>
<p><strong>Accessibility first.</strong> Voice in, voice out. You can ask in Spanish, Somali, Arabic, Polish and other languages and still get the right local answer. And yes, there's a Bristolian dialect mode.</p>
<p><strong>Hyperlocal, city by city.</strong> Share your location (or don't), and Nyata works out which area you're in. It's built so someone in one city never gets another city's services. If there's nothing listed near you yet, the goal is for it to say so plainly rather than make something up.</p>
<p><strong>Structured local data.</strong> A charity listing isn't a paragraph of text. Each session has a type, a day, times, an address and a capacity, so the AI answers from facts, not guesses.</p>
<p><strong>An API that knows its place.</strong> Nyata has a public API and an MCP server spec with seven tools, so other apps and AI agents can ask "what's near here?" One decision I'm proud of: the public API is scoped to <em>discovery</em>. Ask it to complete a booking and it tells you where to go, rather than taking the transaction itself. The taxi firms, salons and charities on Nyata own their customer relationships. An API that took those transactions away would be competing with the very partners the platform depends on.</p>
<h2>What I'll be writing about</h2>
<p>This blog is the engineering side of building a hyperlocal AI product, including the parts that don't go to plan. Some of what's coming:</p>
<ul>
<li><strong>Forcing an app update without shipping an app update.</strong> How we make every phone on iOS and Android move to the latest version, and how to do it without locking anyone out.</li>
<li><strong>Your profile photo knows where you live.</strong> Phone photos carry GPS data. Here's how we strip it on upload, on the phone and again on the server.</li>
<li><strong>Notifications that say <em>where</em>, not <em>what</em>.</strong> A small glass bubble that tells you there's something new in private messages, community chat or the staff hub, without exposing names or message text on screen.</li>
<li><strong>The discovery-only API.</strong> The full reasoning, and what we gave up.</li>
<li><strong>Glass that you can actually read.</strong> Building a translucent "liquid glass" UI where legibility wins every argument.</li>
<li><strong>Multilingual in, local out.</strong> Handling questions in many languages against data that lives in one.</li>
</ul>
<h2>Follow along</h2>
<p>If you're building something local, working on LLMs with real-world data, or care about making technology useful for people who are usually an afterthought, I'd love you to follow along. Questions and pushback in the comments are very welcome.</p>
<p>You can try Nyata at <strong><a href="https://nyataai.co.uk">nyataai.co.uk</a></strong>, and find us on <a href="https://www.linkedin.com/company/nyataai/">LinkedIn</a> and <a href="https://www.instagram.com/nyata_app/">Instagram</a>.</p>
<p>Next up: forcing app updates across two app stores, and the one setting that could have locked every user out.</p>
<p><em>Ibrahim, founder of Nyata</em></p>
]]></content:encoded></item></channel></rss>