<?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" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Mikebraid on Medium]]></title>
        <description><![CDATA[Stories by Mikebraid on Medium]]></description>
        <link>https://medium.com/@mikebraid60?source=rss-aafdf00da717------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*chdgZc5odFkraBbV</url>
            <title>Stories by Mikebraid on Medium</title>
            <link>https://medium.com/@mikebraid60?source=rss-aafdf00da717------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 22:54:03 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mikebraid60/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Covertor — Art of Transformation]]></title>
            <link>https://medium.com/@mikebraid60/covertor-art-of-transformation-3694776ad3fb?source=rss-aafdf00da717------2</link>
            <guid isPermaLink="false">https://medium.com/p/3694776ad3fb</guid>
            <category><![CDATA[convertor]]></category>
            <category><![CDATA[document]]></category>
            <category><![CDATA[pdf-to-word]]></category>
            <category><![CDATA[documentconversion]]></category>
            <dc:creator><![CDATA[Mikebraid]]></dc:creator>
            <pubDate>Tue, 11 Aug 2026 19:31:32 GMT</pubDate>
            <atom:updated>2026-08-11T19:34:25.875Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qjqoqTU-uW0cA0OXZNGGkQ.png" /></figure><h3>Building a Document Converter That Never Stores Your Documents</h3><h3>What it actually takes to make “we don’t keep your files” a true statement, not a privacy-policy sentence</h3><p>Every free online file converter says some version of the same thing: <em>“your privacy matters to us.”</em> Almost none of them show you what that actually means in the infrastructure. It’s easy to write a sentence in a privacy policy. It’s a different thing entirely to architect a system where that sentence is structurally true — where there is no code path, no database table, no forgotten backup, that could hold onto your document even if someone wanted it to.</p><p>That was the constraint I set for <a href="https://con.vertor.app/">Vertor</a>, a free Word-to-PDF and PDF-to-Word converter: <strong>no document is ever written to persistent storage.</strong> Not “deleted after 30 days.” Not “encrypted at rest, indefinitely.” Gone, within minutes, as a property of the architecture rather than a policy someone has to remember to enforce.</p><p>Here’s how it’s actually built, and why some of the more interesting decisions came from problems I didn’t expect to hit.</p><h3>The shape of it</h3><p>Vertor runs entirely on AWS, entirely serverless — nothing sitting idle waiting for traffic, nothing to patch, nothing to scale manually.</p><pre>Browser<br>  │<br>  ▼<br>CloudFront  (one domain, so the API is same-origin with the site)<br>  ├── /*         → S3            static frontend<br>  └── /api/*     → API Gateway / Lambda Function URLs<br>                     ├── presigned upload URL<br>                     ├── the actual conversion<br>                     └── anonymous usage analytics</pre><pre>S3           →  files exist here for minutes, not indefinitely<br>DynamoDB     →  anonymous event counts only, never document content<br>Cognito      →  admin dashboard authentication</pre><p>Static assets — HTML, CSS, JS — sit in a private S3 bucket, served through CloudFront. Nothing there is interesting from a security standpoint; it’s the API side where the real design decisions live.</p><h3>The part that’s actually hard: making the conversion good</h3><p>A lot of “free PDF converter” tools cut a corner most people never notice until it bites them: they use a simplified parser that <em>approximates</em> a document’s layout instead of genuinely rendering it. That works fine for a page of plain paragraphs. It falls apart the moment a document has a merged table cell, a section break that changes page orientation halfway through, or a font that isn’t installed on whatever server is doing the converting.</p><p>Vertor uses an actual document layout engine — the same category of software that renders a document when you print it from Word itself — running headless inside a Lambda container image. That engine is genuinely heavy: the container image is the better part of a gigabyte once you include the fonts needed to render documents accurately. That single fact shaped almost every other architectural decision downstream of it.</p><h3>Why a “simple” file upload needed three steps instead of one</h3><p>The obvious design — browser uploads a file directly to an API endpoint, gets a <a href="https://con.vertor.app/">converted file</a> back — runs into a wall almost immediately: <strong>AWS Lambda has a hard 6MB payload limit on synchronous invocations</strong>, full stop, regardless of which service sits in front of it. A Word document with a few embedded images blows past that without trying.</p><p>The fix is the standard pattern for this class of problem, and it’s a nice piece of architecture once it clicks: the browser never sends the file to Lambda at all.</p><ol><li>The browser asks a small Lambda function for a <strong>presigned S3 URL</strong> — a temporary, scoped credential that’s good for exactly one upload, to exactly one location, for five minutes.</li><li>The browser uploads the file <strong>directly to S3</strong>, bypassing Lambda’s payload limit entirely.</li><li>A second Lambda — the one running the actual conversion engine — picks the file up from S3, converts it, writes the result back, and deletes the original the instant it’s done.</li></ol><p>That last step is where “no storage” actually gets enforced, not just promised: the source file’s lifetime in S3 is measured in seconds, deleted programmatically the moment conversion succeeds (or fails). A storage lifecycle rule sits underneath as a backstop — anything left in the bucket for any reason auto-expires within a day — but in the normal path, nothing survives long enough for that rule to ever matter.</p><h3>The bug that taught me the most</h3><p>Partway through building this, <a href="https://con.vertor.app/">PDF-to-Word conversions</a> started failing with no useful error. Word-to-PDF worked fine; the reverse direction didn’t. The cause, once I found it, was a genuinely interesting one: <strong>the conversion engine’s default PDF import opens files as a <em>drawing</em>, not as text</strong> — it treats each page as a canvas of positioned shapes, which is a completely reasonable default for viewing a PDF, and a completely unusable one for exporting to an editable Word document, since you can’t export a drawing to a text-flow format at all.</p><p>The fix was a single command-line flag forcing the <em>text-reconstruction</em> import path instead of the default one — but finding that took actually understanding what the tool was doing under the hood, not just calling it and hoping. That’s the kind of bug that never shows up in a demo with a simple test file, and exactly the kind that erodes trust in a tool the first time a real user hits it.</p><p>A second one, smaller but sharper: <strong>API Gateway’s HTTP APIs have a fixed 30-second timeout with no way to raise it.</strong> A cold-started conversion — pulling a large container image, starting a heavyweight rendering engine, then actually converting a document — can genuinely take longer than that on a first request. The fix was routing the conversion endpoint around API Gateway entirely, straight from CloudFront to a Lambda Function URL, which isn’t bound by that same ceiling. Small architectural detail, but the difference between “works in testing” and “works under real conditions.”</p><h3>Security, as a property of the system rather than a promise</h3><p>A few specific decisions, and why each one matters:</p><p><strong>Nothing is publicly reachable except through the front door.</strong> The S3 buckets, the conversion function — none of it has a public endpoint of its own. Every path in runs through CloudFront’s Origin Access Control, which cryptographically signs requests so that only <em>this specific distribution</em> can reach the backend, not “anyone who finds the URL.”</p><p><strong>Permissions are scoped per function, not shared.</strong> The function that generates upload URLs can write to S3. The function that serves analytics can read (only) from the events table. Neither can do the other’s job. If any one piece were ever compromised, the blast radius is deliberately small.</p><p><strong>The admin dashboard is genuinely gated, not security-through-obscurity.</strong> It sits behind real authentication (Cognito), and the API it calls independently re-checks group membership server-side — not just “is this person logged in,” but “is this person specifically authorized,” checked twice, in two different places, on purpose.</p><p><strong>Analytics are anonymous by construction.</strong> The dashboard shows conversion counts, error rates, visit numbers — never a filename, never file content, never anything that could re-identify a specific document. It’s not that identifying data gets scrubbed after collection; it’s never collected in the first place.</p><h3>Why this matters beyond one tool</h3><p>None of this is exotic. It’s mostly disciplined use of ordinary building blocks: presigned URLs, scoped IAM permissions, lifecycle rules, origin access control. What’s uncommon isn’t the individual techniques — it’s applying all of them consistently, to a category of product (a free file-conversion tool) where it would have been easy to cut every one of these corners and nobody would have noticed until something went wrong.</p><p>“We don’t store your files” is a sentence any privacy policy can contain. Whether it’s <em>true</em> is a question of architecture, not wording — and it’s worth asking, the next time you upload something sensitive to a free tool, whether anyone actually built it that way.</p><p><em>Vertor is free to use at </em><a href="https://con.vertor.app/"><em>con.vertor.app</em></a><em> — Word to PDF and PDF to Word, no sign-up, no watermark, and now you know exactly what happens to your file after you hit convert.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3694776ad3fb" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>