<?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 Andrew Sokolov on Medium]]></title>
        <description><![CDATA[Stories by Andrew Sokolov on Medium]]></description>
        <link>https://medium.com/@andrewsokolov?source=rss-3df6cc63a371------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Fg2gFAv0OguwgxkGUWfS-g.png</url>
            <title>Stories by Andrew Sokolov on Medium</title>
            <link>https://medium.com/@andrewsokolov?source=rss-3df6cc63a371------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 23:54:50 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@andrewsokolov/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[I Could Solve the Problem, but I Could Not Explain It in English. That Is How ExtraBrain Started]]></title>
            <link>https://medium.com/@andrewsokolov/i-could-solve-the-problem-but-i-could-not-explain-it-in-english-that-is-how-extrabrain-started-e7b88558d472?source=rss-3df6cc63a371------2</link>
            <guid isPermaLink="false">https://medium.com/p/e7b88558d472</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[interview]]></category>
            <category><![CDATA[generative-ai-tools]]></category>
            <dc:creator><![CDATA[Andrew Sokolov]]></dc:creator>
            <pubDate>Sun, 07 Jun 2026 11:46:34 GMT</pubDate>
            <atom:updated>2026-06-07T11:46:34.504Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*09kCoFN44jTGn0izdb3EEA.png" /></figure><p>After relocating, I learned pretty quickly that technical interviews abroad are not just about algorithms, architecture, and experience. They are also a small theatrical performance in English, where you are simultaneously a candidate, translator, tech lead, part-time stand-up comedian, and a person trying to remember why they opened Zoom in the first place.</p><p>I walked out of several interviews with a very strange feeling. I knew the topic. I had worked with similar problems before. I could explain why a queue was needed here, why consistency mattered there, and why turning everything into microservices just because the word sounds serious was probably not the best idea.</p><p>But in the moment my brain sometimes said: “Good luck, I’m on vacation.”</p><p>That is how I started building ExtraBrain, a macOS app that helps not after the call, but while the conversation is actually happening.</p><p>Who is writing: Andrii, software developer and indie hacker. I am building ExtraBrain, a desktop AI assistant for interviews, live coding, system design, and difficult technical calls.</p><h3>After Relocation, Interviews Became a Different Sport</h3><p>Before relocation, I thought I more or less understood how technical interviews worked. There is a question, there is your experience, and there is an answer. Sometimes there is a whiteboard, sometimes a code editor, sometimes an interviewer with the facial expression of someone who has already spoken to seventeen candidates that day asking about algorithmic complexity.</p><p>Abroad, everything became more interesting. A different market. A different rhythm. Different small talk. You need to switch to English quickly, keep the question in your head, remember the context from the previous ten minutes, look at the code, avoid losing your train of thought, and still not look like a person who joined the call by accident while trying to order sushi.</p><p>The most frustrating part was that the problem was often not knowledge. I did not leave interviews thinking, “Wow, I had no idea what cache invalidation was.” More often it was: “Why did I explain that so awkwardly? Why did I forget the edge case? Why didn’t I ask the obvious clarification? Why did the good answer arrive exactly twelve minutes after the call ended?”</p><p>I think many people have been there. Especially after moving to another country, when everything around you is already new: documents, banks, rent, taxes, random household quests. And then, on top of that, you are expected to cheerfully explain distributed systems as if you did not sleep four hours because the washing machine in your new apartment sounded like an emergency alarm.</p><h3>A ChatGPT Tab Was Not Enough</h3><p>At first, I tried to help myself with a regular AI chat. Open a tab, paste part of the question, ask it to structure the answer. On paper, it sounds great.</p><p>In a real interview, it looked different. By the time you copy the question, write a prompt, explain the context, paste a piece of code, and wait for the answer, the moment is already gone. The interviewer is looking at you. You are looking back. Somewhere in another browser tab, a beautiful suggestion is being born one and a half minutes too late.</p><p>I realized I did not need another chat where I had to manually bring the entire world into the prompt. I needed a tool that was already nearby: something that could listen to the conversation, understand the context, look at the screen, and quickly suggest which direction to think in.</p><p>Not “answer instead of me.”</p><p>More like: “Hey, don’t forget to clarify this constraint.” “There is a trade-off here.” “Start by explaining the structure.” “This sounds less like a coding question and more like a design question.”</p><h3>What ExtraBrain Does</h3><p>ExtraBrain is a macOS app for situations where you need to think right now.</p><p>I start a session before an interview or a technical call. The app listens to the microphone and system audio, builds a transcript, and can take into account what is visible on the screen: code, browser, diagram, IDE, or document.</p><p>When a difficult question comes up, you can quickly ask for analysis. For example:</p><ul><li>what am I actually being asked right now;</li><li>how should I structure the answer;</li><li>which edge cases should I mention;</li><li>what should I clarify before answering;</li><li>which risks and trade-offs should not be forgotten;</li><li>how can I explain the solution briefly;</li><li>where might I already be going in the wrong direction.</li></ul><p>After the call, you can review the history and summary, but that is not the main point. The main idea of ExtraBrain is help in the moment. Not an hour later, when you finally come up with the perfect answer in the shower, but right when that answer can still save the conversation.</p><h3>Who It Is For</h3><p>Right now, I see several use cases:</p><ul><li>technical interviews;</li><li>live coding;</li><li>system design interviews;</li><li>architecture calls;</li><li>debugging and review sessions;</li><li>product or customer calls;</li><li>meetings where you need to catch details quickly and ask the right questions.</li></ul><p>I think it can be especially useful for people after relocation or for anyone who often works in a non-native language. Sometimes you know the answer but get stuck on the wording. Sometimes you understand the architecture problem but do not have time to lay it out clearly. Sometimes you just need a small push so your brain stops pretending to be a loading spinner.</p><h3>This Is Not a Cheat Sheet</h3><p>Yes, I understand the topic is sensitive. AI during interviews immediately raises the question: where is the line between support and cheating?</p><p>I do not want to position ExtraBrain as an invisible cheat sheet that makes a candidate look better than they really are. I am much more interested in scenarios where using tools is transparent, allowed, or at least openly discussed.</p><p>The world is already changing. Developers use AI in their IDEs, at work, in documentation, during code review, and while debugging. The question is not whether AI will disappear from engineering workflows. It will not. The question is what rules we build around it.</p><p>For me, ExtraBrain is closer to a second monitor with notes, a good interviewer who asks a guiding question, or a colleague who helps you not forget an important nuance. It does not replace experience. If a person does not understand the topic, a hint will not magically turn them into a senior engineer.</p><p>But if a person is nervous, switching between languages, or losing context, a small push can help show their real level instead of just measuring stress.</p><h3>Where the Project Is Now</h3><p>ExtraBrain is currently in beta. It is a macOS application, and I am actively improving it.</p><p>I am especially interested in feedback from people who have gone through interviews abroad, conducted technical interviews, worked with candidates, or simply spend a lot of time on complex engineering calls.</p><p>You can download it here:</p><p><a href="https://extrabrain.app/download/">Download ExtraBrain for Mac - Free AI Interview &amp; Meeting App | ExtraBrain</a></p><p>Feedback is welcome at:</p><p><a href="/service/mailto:team@extrabrain.app">team@extrabrain.app</a></p><p>or through the form on the website.</p><p>I am particularly interested in three questions:</p><ul><li>in which situations does this kind of tool feel useful and acceptable;</li><li>where do you personally draw the line between support and cheating;</li><li>what would make an assistant like this useful not only for interviews, but also for everyday engineering work.</li></ul><p>In short: I started building ExtraBrain because after relocation I personally felt how much interviews can test not only knowledge, but also stress tolerance, language, reaction speed, and the ability to assemble your thoughts on the fly.</p><p>I wanted to build a tool that does not pass the interview instead of a person, but helps the person not lose their own brain at the worst possible moment.</p><p>And if the project does not take off, at least I will have an app that reminds me: sometimes the best pet project starts with the sentence, “Okay, I am not failing the next interview like that again.”</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e7b88558d472" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I Finally Stopped Overthinking Every Slack Message]]></title>
            <link>https://medium.com/@andrewsokolov/how-i-finally-stopped-overthinking-every-slack-message-540ee0d4bebb?source=rss-3df6cc63a371------2</link>
            <guid isPermaLink="false">https://medium.com/p/540ee0d4bebb</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[ai-assistant]]></category>
            <category><![CDATA[interview]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Andrew Sokolov]]></dc:creator>
            <pubDate>Wed, 06 May 2026 15:39:18 GMT</pubDate>
            <atom:updated>2026-05-06T15:39:18.041Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Yt4WbPb7tQDytjyFCWpnWQ.jpeg" /></figure><p>We’ve all been there: staring at a draft in Slack for ten minutes, wondering if our tone sounds too blunt or if that preposition is actually doing its job. When you’re working in a high-stakes English-speaking environment, that “send” button can feel like a huge hurdle.</p><p>I recently started using <strong>ExtraBrain</strong>, and it’s shifted my workflow from “anxious editor” to “confident communicator” almost overnight. Here is how I’ve been using one of its most powerful features <strong>Custom Profiles</strong> to level up my professional game.</p><h3>The Power of “Better Writing” Profile</h3><p>The beauty of ExtraBrain isn’t just that it’s smart; it’s that you can train it to think exactly like you want. I created a custom profile and named it <strong>“Better Writing.”</strong> Inside this profile, I added a specific set of rules. I didn’t want generic AI fluff; I wanted my own thoughts polished. My prompt tells the app to take my draft and reconstruct it with professional syntax while preserving the original intent.</p><pre>##Role<br>You are a professional communications assistant specialized in high-level corporate English. Your job is to refine draft messages to sound natural, authoritative, and clear.<br><br>##Context<br>The input will be a screenshot of a draft message, likely from a Slack or email interface. The user is a professional working in an English-speaking environment who wants to ensure their syntax and tone are perfect before sending.<br><br>##Instructions<br>Analyze the text within the screenshot.<br><br>Correct all grammatical errors and awkward phrasing.<br><br>Optimize for a &quot;Professional Tech&quot; tone  clear, polite, and direct.<br><br>Ensure the sentence constructions sound native and fluid.<br><br>Preserve the original intent and any technical jargon used.<br><br>##Constraints<br>Do not use &quot;fluff&quot; or overly formal &quot;Victorian&quot; English.<br><br>Keep the length similar to the original draft.<br><br>Do not include any meta-talk like &quot;Here is your refined message.&quot;<br><br>##Output<br>Return only the corrected, ready-to-paste text.</pre><p><strong>Pro Tip:</strong> I also toggled on the <strong>“Start analysis after region screenshot”</strong> setting. This removes a step from the process, making the transition from “thought” to “result” nearly instant.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o98TIpoZ7jNRE8yh7Rs4Qw.png" /></figure><h3>From Draft to “Sent” in Two Clicks</h3><p>Here is exactly how I use it during a typical workday when I need to reply to a teammate:</p><ol><li><strong>Drafting:</strong> I type my rough message directly into the Slack text box, just getting my ideas down without worrying about perfect grammar.</li><li><strong>Capturing:</strong> Instead of copying and pasting text into another window, I just <strong>right-click the Capture button</strong> in ExtraBrain. This lets me select a specific area of my screen.</li><li><strong>Analyzing:</strong> I draw a box over my Slack message. Because of my “Auto-start” setting, ExtraBrain immediately reads the text in that screenshot and applies my “Better Writing” rules.</li><li><strong>Perfecting:</strong> Within seconds, the analysis window shows a refined, more professional version of my message.</li></ol><h3>The Result</h3><p>The difference is night and day. I simply copy the new version, paste it back into Slack, and hit send. The text constructions are native-level, the tone is sophisticated, and most importantly, I don’t feel like I’m “pretending” I’m just communicating at the speed of my own ideas.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q3bIdh_CybkCPbhzgE1IdA.png" /></figure><p>If you’re looking for a way to bridge the gap between your thoughts and a professional English output without the constant second-guessing, setting up a custom profile on <a href="http://extrabrain.app/"><strong>extrabrain.app</strong></a> is a total game-changer.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=540ee0d4bebb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I Got a Powerful AI Brain for Free — Using Kimi K2.6 on ExtraBrain via Fireworks.ai]]></title>
            <link>https://medium.com/@andrewsokolov/how-i-got-a-powerful-ai-brain-for-free-using-kimi-k2-6-on-extrabrain-via-fireworks-ai-3596d7644d18?source=rss-3df6cc63a371------2</link>
            <guid isPermaLink="false">https://medium.com/p/3596d7644d18</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[fireworks]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[kimi-k2]]></category>
            <category><![CDATA[knowledge-management]]></category>
            <dc:creator><![CDATA[Andrew Sokolov]]></dc:creator>
            <pubDate>Sat, 25 Apr 2026 21:39:38 GMT</pubDate>
            <atom:updated>2026-04-26T18:56:17.210Z</atom:updated>
            <content:encoded><![CDATA[<p>When I first downloaded <a href="https://extrabrain.app/">ExtraBrain</a>, I was excited about the idea of having a personal AI-powered knowledge companion. But like most people, I paused when I saw “connect your AI provider.” I didn’t want to commit to another subscription or burn through an API budget while just exploring.</p><p>Then I discovered Fireworks.ai — and its <strong>$5 free credit</strong> for new users. That was more than enough to get started with Kimi K2.6, one of the most capable open-source models available right now. Here’s exactly how I set it up.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*34pK2HyD8cNWEz6rYbjBgA.png" /></figure><h3>Why Kimi K2.6 on Fireworks?</h3><p>Before we dive in, a quick word on why this combination is worth your time.</p><p><strong>Kimi K2.6</strong> is Moonshot AI’s flagship model — a multimodal, agentic model with a 256k token context window that genuinely rivals closed-source models like Claude Opus and Gemini Pro. It’s fast, capable, and excellent at reasoning, document analysis, and multi-turn conversations.</p><p><strong>Fireworks.ai</strong> is Moonshot’s preferred launch partner for Kimi models. They offer the fastest inference endpoint (up to 200 tokens/second), a fully OpenAI-compatible API, and — crucially — <strong>$5 in free credits</strong> when you sign up. That’s real money you can use immediately.</p><p>Combined with ExtraBrain’s support for custom LLM providers, you can have a premium-grade AI second brain running in under 10 minutes, for free.</p><p>Let’s do it.</p><h3>Step 1: Register on Fireworks.ai and Claim Your $5</h3><p>Head to <a href="https://fireworks.ai/">fireworks.ai</a> and create a free account. The signup process is straightforward — email, password, done.</p><p>Once you’re in, you’ll see <strong>$5 in free credits automatically added to your account</strong>. No credit card required to get started.</p><p>Take a moment to browse the model library. You’ll find Kimi K2.6 listed prominently — it’s one of their flagship offerings. Note the model ID, you’ll need it shortly:</p><p>accounts/fireworks/models/kimi-k2p6</p><p>The base URL for all Fireworks API calls (OpenAI-compatible) is:</p><p><a href="https://api.fireworks.ai/inference/v1">https://api.fireworks.ai/inference/v1</a></p><p>Keep this tab open — you’ll come back to it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xL9ftQLB7z0003atd4aGww.png" /></figure><h3>Step 2: Download ExtraBrain</h3><p>Go to <a href="https://extrabrain.app/">extrabrain.app</a> and download the app for your platform. ExtraBrain is available on macOS, iOS, and other platforms.</p><p>Install it as you normally would. When you open it for the first time, you’ll be taken through an <strong>onboarding flow</strong> — this is where the magic happens</p><h3>Step 3: Choose “Custom LLM Provider” During Onboarding</h3><p>During the onboarding setup, ExtraBrain will ask you how you want to power your AI. You’ll see options for popular providers like OpenAI, Anthropic, and others.</p><p>Look for the <strong>“Custom LLM Provider”</strong> option and select it. This is the key that unlocks the ability to bring your own model from any OpenAI-compatible API — including Fireworks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Lz-xZEjmyi8JrJ-qonCXQw.png" /></figure><h3>Step 4: Enter the Base URL and Model ID</h3><p>With the custom provider option selected, ExtraBrain will ask you for two pieces of information:</p><p><strong>Base URL:</strong></p><p><a href="https://api.fireworks.ai/inference/v1">https://api.fireworks.ai/inference/v1</a></p><p><strong>Model ID:</strong></p><p>accounts/fireworks/models/kimi-k2p6</p><p>Enter both exactly as shown above. The model ID format is specific to Fireworks — it follows the accounts/fireworks/models/{model-name} pattern.</p><p>If you want the even faster Kimi K2.5 Turbo variant (especially useful for agentic workflows), you can alternatively use:</p><p>accounts/fireworks/routers/kimi-k2p5-turbo</p><h3>Step 5: Generate Your API Key on Fireworks</h3><p>Now switch back to your Fireworks tab. In the Fireworks dashboard, navigate to:</p><p><strong>Account → API Keys</strong> (or go directly to app.fireworks.ai/api-keys)</p><p>Click <strong>“Create API Key”</strong>, give it a name like ExtraBrain, and copy the key. It will look something like:</p><p>fw-xxxxxxxxxxxxx</p><blockquote><em>⚠️ </em><strong><em>Important:</em></strong><em> Copy this key immediately and store it somewhere safe. Fireworks only shows it once.</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8iOD2axidkZHIVIHb8d0OQ.png" /></figure><h3>Step 6: Paste the API Key into ExtraBrain</h3><p>Back in ExtraBrain, paste your Fireworks API key into the API key field to complete the setup.</p><p>Hit <strong>Connect</strong> (or whatever the confirmation button is in your version), and ExtraBrain will verify the connection. If everything is entered correctly, you’ll see a success confirmation and land in your new, Kimi-powered workspace.</p><p>That’s it. You’re done.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IlS3bjcGmlplOwojxTSzgg.png" /></figure><h3>What You Now Have</h3><p>With this setup, you’re running <strong>Kimi K2.6</strong> — a model with a 256,000 token context window — inside ExtraBrain. You can feed it entire PDFs, lengthy notes, and complex queries, and it will keep up.</p><p>Your $5 of free Fireworks credits will go a surprisingly long way. Kimi K2.6 costs $0.60 per million input tokens and $3 per million output tokens. A typical conversation costs a few cents at most.</p><h3>Troubleshooting Tips</h3><ul><li><strong>Connection fails?</strong> Double-check that the base URL ends in /v1 and that the model ID matches exactly (it&#39;s case-sensitive).</li><li><strong>API key not working?</strong> Make sure you copied the full key and didn’t accidentally trim any characters.</li><li><strong>Want a different model?</strong> Fireworks supports dozens of other models — DeepSeek V3, Llama 3.3, Qwen 3, and more. Just swap the model ID in ExtraBrain’s settings.</li></ul><h3>Final Thoughts</h3><p>I was genuinely surprised how smooth this was. ExtraBrain’s custom provider support means you’re never locked into a single AI vendor, and Fireworks’ free credits make it risk-free to experiment.</p><p>If you’ve been on the fence about setting up a proper AI second brain, this is your sign to try it. The barrier is lower than you think — it takes less time than brewing a coffee.</p><p>Give it a shot, and let me know how it goes in the comments.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3596d7644d18" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>