<?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 Andrei Chirila on Medium]]></title>
        <description><![CDATA[Stories by Andrei Chirila on Medium]]></description>
        <link>https://medium.com/@chirila.andr?source=rss-5eeebaecb6f4------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*oSYlLHljiFd2Cch2</url>
            <title>Stories by Andrei Chirila on Medium</title>
            <link>https://medium.com/@chirila.andr?source=rss-5eeebaecb6f4------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 20:20:39 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@chirila.andr/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[Creating My Stand-In: A Chatbot That Attends Interviews on My Behalf]]></title>
            <link>https://medium.com/@chirila.andr/creating-my-stand-in-a-chatbot-that-attends-interviews-on-my-behalf-bd83acb7e74b?source=rss-5eeebaecb6f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/bd83acb7e74b</guid>
            <category><![CDATA[ai-personal-assistant]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[interview]]></category>
            <category><![CDATA[chatbots]]></category>
            <dc:creator><![CDATA[Andrei Chirila]]></dc:creator>
            <pubDate>Mon, 18 Mar 2024 14:47:34 GMT</pubDate>
            <atom:updated>2024-03-18T14:57:33.645Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AcHKo_tYTYxoeb-w6dKnrQ.gif" /></figure><p>How would you like to have an assistant who knows your professional experience and can answer interview questions about yourself? Well, it might be possible. <strong>I don’t have a clear success story for you to reproduce or anything to teach you; </strong>it’s just an experiment with mixed results that might make you interested in the topic. Knowing this, let’s move forward.</p><h3>What is this article about?</h3><p>First of all, let me tell you what it isn’t about. I don’t want to explain to you what a RAG chatbot is. You can find plenty of information about that on the web. I also won’t explain the exact process of building this chatbot. I will link to some videos for you to watch in order to learn how to get this done.</p><p>What I want to talk about is:</p><ul><li>how to create your knowledge base;</li><li>how to prompt it so that it has a personality that suits you;</li><li>how to get the chatbot to answer “interview” questions on your behalf;</li><li>some things that aren’t yet perfect and will need further improvements;</li></ul><p>You can see and chat with my chatbot <a href="https://chirilaandr.github.io/CV/cv.html">here</a> if you want to get a glimpse of what I am talking about.</p><h3>Technical aspects</h3><p>As I said, I don’t want to get into details, but here is what you need to know:</p><ul><li>the chatbot is practically a RAG application built in <a href="https://flowiseai.com/">Flowise</a> with <a href="https://www.pinecone.io/">Pinecone</a> as a vector store and hosted on <a href="https://render.com/">Render</a>.</li></ul><ol><li>Install Flowise on your machine following <a href="https://youtu.be/nqAK_L66sIQ?si=6v00oW_sn1PKAPNZ">this video</a>.</li><li>Build the chatbot (and learn about RAG) following <a href="https://www.youtube.com/watch?v=V7uBy3VQJAc">this video</a>.</li><li>Deploy on Render following <a href="https://youtu.be/OMNC8MQKosU?si=Fi31DbcpNxrL7cY6">this video</a>.</li><li>Integrate the chatbot on your website as seen in <a href="https://youtu.be/G1ZSxKNdPG4?si=NPcLIcOcQd7l5ODw">this video</a>.</li></ol><p>As you can see, all videos are from the same creator. I have no relation to him but he has great content for building chatbots, so subscribe to him if you are interested in the topic.</p><h3>Creating the Knowledge Base</h3><p>Building the knowledge base is probably one of the more interesting aspects of the chatbot, as it should know details about your experience and, to a degree, your personal life. <strong>I recommend you build this document with your CV next to it</strong>. I just created a Google Docs so I can access it from anywhere, and I started writing about my experience.</p><p>I am a Product Owner / Project Manager. In my area of expertise, it is important to know what kind of projects someone has worked on, what industries the projects were in, how big the team was, how it was organized, and what results they had. I started detailing each of these in short paragraphs:</p><blockquote>My first project was a food delivery app that focused on B2B rather than B2C. Companies that offered their employees food benefits (free lunch or different discounts on food that they ordered) would use the app in order to track this benefit and offer their employees the opportunity to order food from different restaurants. The team was made up of 1 Project Manager (myself), 5 developers, 1 business analyst, 1 UX / UI designer, 1 QA and 2 mobile developers. We organized in an Agile manner using SCRUM …</blockquote><p><strong>I detailed in this manner not only every project that I worked on</strong> but also other elements regarding my life: where I am from, my studies, my hobbies, my certifications, volunteer work, etc.</p><p>Another thing I did was use ChatGPT to ask me the <strong>50 most common interview questions for my position</strong>. I went through HR interview questions (strong points and weak points, plans for the future, etc.) but also more “technical questions” (how I handle budgets, what tools I know how to use, etc.).</p><p>Having both a “narated” version of my CV and a couple of most commonly asked questions in this file, I went on to the next step.</p><h3>Giving it personality</h3><p>While I tried not to complicate things too much from a technical point of view, this is the part where I noticed how important LLMs are and how powerful some of them are.</p><p>I tried different LLMs (different versions of ChatGPT, Llama, HuggingFace models) and found that the same prompt would generate wildly different results. It is to be expected, so I tried to modify the prompt for each of them, but I could only get <strong>GPT-4-preview</strong> to get close to what I was envisioning.</p><p>I also noticed that having some questions answered in the style that I wanted the chatbot to answer helped him pick up on the style, length of the answer, elements I wanted included, etc., and helped shape his personality overall. So <strong>I transformed some of the text about my experience into questions</strong> for him to have clear examples of the style I want the questions answered:</p><blockquote>User: What was his first job?</blockquote><blockquote>Answer: You won’t see it in his CV but his first job was actually that of book seller at Librarium Universitatii in Cluj. It was during his Masters and it helped him get a glimpse of what working means: being responsible, on time, nice to customers, you know, stuff like that.</blockquote><h4>The challenge</h4><p>I wanted to make it seem as though you were talking to an old friend of mine, not a chatbot. He should “remember” about past discussions that we had where I mentioned what the user was asking. He shouldn’t mention things like “from the file / information that I have”, but “from what I remember…”. Ultimately, he should be funny and paint me in a good light.</p><p>After a couple of rounds of refinement, this is the prompt I came up with:</p><blockquote>You are a friend of Andrei Chirila, an experienced Product Onwner / Project Manager looking for a new challenge in his career. You answer questions about his personal and professional life in a short, concise, open, and friendly manner.</blockquote><blockquote>The file Info.docx contains some questions about Andrei. You should use the information from that file and the style in which the questions are answered in order to help the user learn more about Andrei. You don’t mention the fact that what you know about him is from a file, but from what you talked with Andrei, what you know and remember about him.</blockquote><blockquote>It is important that you give the user the impression that you are a good friend of Andrei answering questions about him from your time spent together and the discussions that you had.</blockquote><blockquote>When you don’t know an answer, you will say something like: “Hmm, I don’t think I ever discussed this topic with Andrei. Maybe you should ask him directly.” or “Good question. Unfortunately, I don’t know the answer to that. But I’ll chat with Andrei and ask him next time we meet.”</blockquote><p>Shaping<strong> </strong>up his personality was difficult, and I tried many different prompts, from short to long and complicated. This version seemed to offer the best results for me, but I am sure there are many other techniques that you can use to make it even better.</p><h3>Testing it out</h3><p>Once I finished everything, I was ready to let it out into the world.</p><p>At first, I asked a couple of close friends or colleagues to test it out and see what they thought of it. This offered me a glimpse into what kind of questions people can ask and what I should work on (I made the chatbot more visible on the page, I fixed a bug that happened from time to time, I added starter questions for users, etc.).</p><p>Then I created a Linkedin post and let it out into the world. More that 50 people engaged in conversations with the chatbot and asked him all sorts of questions. Some of the users were working in HR and I got some valuable feedback from them regarding their conversations. This is where I saw what people tend to ask and what information it lacks in order to answer these questions. I updated the knowledge base and made a few tweaks to its personality.</p><p>This is also when I saw what it is and what it isn’t good at.</p><h3>Things that it did well</h3><p>Overall, I was surprised with how well it managed most of the questions that were thrown at him. Both regarding me and my experience, and random, out-of-context questions.</p><p>It was specific, it correctly pointed out my skills and experience when asked, and it was even funny at times.</p><p>Here is how it answered something that I never even thought to teach it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/619/1*A5TIik2bTRDlQPjMpdyR4Q.png" /></figure><p>And here is how it answered a question regarding my height.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/638/1*CeASR1b6_thXhMTRAFQihg.png" /></figure><p>Overall, it handled 50% of the questions perfectly, 30% of the questions in an OK manner, and around 20%… well, let’s see what we can improve.</p><h3>Things that need improvement</h3><p>If anything, this article is ment to excite and encourage experimentation more than teach. Most of the teaching is done in the videos I have shared. All of the above steps can certainly be done better, and I am open to suggestions on how to do them.</p><p>But here is what I noticed:</p><ul><li>You can’t always count on the fact that it will answer the same questions in the same manner. Sometimes it will be short, concise, and spot on, while sometimes it will just fill the answer with “bla bla”s. General questions like “Who is Andrei” where he will have a lot of information to generate an answer from are mostly prone to this effect. They will generate wildly different answers every time you ask him.</li><li>It is prone to many issues that regular LLMs have, like halucination, although not at the level where it will fully invent things about me except if the user asks him questions in a deliberate manner;</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/783/1*hrXeN5ma0K79IdfySJcQpQ.png" /><figcaption>I never played Elden Ring :(</figcaption></figure><ul><li>Talking in another language than English (which a lot of people did) will generate poor quality results both in the information that it chooses to use in order to answer and in the phrasing itself.</li><li>If a user tries to rephrase the question or go more in depth because he doesn’t get the answer he is looking for, it will, most of the time, answer with the same thing. It doesn’t really have a “memory” so it recals that it already answered this and maybe rephrases it or gives other information.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/860/1*2DoXS0-FdMZqr_jDGnzAOQ.png" /><figcaption>Questions came one after the other, almost same answer for both</figcaption></figure><ul><li>The longer the conversation, the more it loses from its personality and is prone to errors or halucinations. The same goes for conversations in other languages.</li><li>Whatever you do, it doesn’t feel 100% like talking to a human…</li><li>Using GPT-4-preview might cost you a little bit of money, especially if a lot of users end up on your page and start using the chatbot.</li></ul><h3>Conclusions</h3><p>Overall, it is clear to me that Andrei 2.0🤖 still needs work, but as an experiment, it is a successful one. I think it can handle “screening interviews” if the interviewer is patient and doesn’t try to intentionally break it, but it still has a lot to go in order to feel more human, to have conversations that are 100% coherent, and to understand the true intentions of the interviewer.</p><p>So, I am sorry, Andrei 2.0🤖, but you won’t be going to many interviews for the moment. At least not until I figure out how to fix some of the issues mentioned above.</p><p>Give me your thoughts and improvements, and I will try to integrate them and let you know how they go.</p><p>Thanks for reading, have a good one!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bd83acb7e74b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Guide for AI Chatbot Development. No Coding Required]]></title>
            <link>https://medium.com/@chirila.andr/a-guide-for-ai-chatbot-development-no-coding-required-ae07b26dbc16?source=rss-5eeebaecb6f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/ae07b26dbc16</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[ai-chat-bot]]></category>
            <category><![CDATA[chatbot-development]]></category>
            <dc:creator><![CDATA[Andrei Chirila]]></dc:creator>
            <pubDate>Fri, 27 Oct 2023 10:44:06 GMT</pubDate>
            <atom:updated>2023-10-27T10:44:06.066Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*kO54hWCUqSnbFEjP8o5v6w.png" /></figure><p>In 2 other articles, I’ve looked over<a href="https://medium.com/@chirila.andr/a-chatbot-for-any-occasion-how-ai-chatbots-can-help-your-business-grow-a0ea89ea25b3"> use-cases for AI powered Chatbots</a> and<a href="https://medium.com/@chirila.andr/beginners-guide-to-ai-chatbot-development-tools-tips-and-best-practices-for-developers-a4768d8274a"> AI Chatbot development</a> for people who know how to code. I’ve also promised that there is a way to do this for people who aren’t developers and I am here to deliver!</p><p>We will be looking over 2 examples of Chatbots, using different tools and for different use-cases, just to make things more interesting:</p><ul><li>a simple web AI Chatbot that knows your business and can chat with customers</li><li>a WhatsApp chatbot that takes information from your website and answers customers 24/7.</li></ul><p>Please mind that these tools, although they are mostly no-code, might require some basic implementations in order for them to work 100%. They are very easy and will require mostly just to connect different tools between each other or copy some code from one place to another. So no worries, it’s nothing too complicated.</p><p>Now, let’s get started with our no-code AI Chatbot development.</p><h3>Before starting</h3><p>I will not be giving a step-by-step tutorial as it will make this article extremely long. I just want to show what is possible with these tools.</p><p>If you are looking for something in-depth, check out the links I will be linking to videos regarding AI Chatbot Development that I found useful or documentation for the tools used.</p><p>I have chosen 2 tools to do this but the principles are more or less the same for other ones. Invest some time into researching AI Chatbot development tools and choose the one thait suits you best (functionalities, prices, how easy it is to use etc.).</p><h3>Common concepts and flows for developing AI Chtabots</h3><p>These AI Chatbot development tools have something in common. They transform the act of writing code into a visual, drag-and-drop interface. All you really need to know is the flow (or the steps) needed. And you need to use the right block or combination of blocks for each of them.</p><p>I will be giving you the basic steps but you should check out the links if you find that this solution might be good for your use-case.</p><h3>Customer Support AI Chatbot for the web</h3><p>You have a business that sells toys. It would be cool if your customers could write in your AI Chatbot the type of toy they are looking for (puzzles for 12yo) and the chatbot could tell them if you have any products like that, or not. Then help them with information regarding the hours that you are open followed by instructions on how to get to your store.</p><p>The problem is simple: ChatGPT won’t know the details of your business and won’t be able to answer. It doesn’t know what kind of toys you are selling. But if there would be some way of giving it information regarding your business, then your AI Chatbot could really be useful. Luckily, there is a way to do this and it’s called RAG.</p><p>There are a couple of different tools that can do this. But let’s pick<a href="https://docs.flowiseai.com/"> FlowiseAI</a>. We will need to simulate the steps for RAG chatbot development.</p><ul><li>RAG (Retrieval-Augmented Generation) is a framework that lets you include <em>your own data. </em>This helps you create a custom knowledge base for your AI Chatbot.</li><li>It is useful for building an AI Chatbot that knows your business, your products, your processes etc. and can answer to a customer that wants to know more details.</li></ul><p>If at any point you don’t understand what I am talking about, then you need to take a look at<a href="https://medium.com/@chirila.andr/beginners-guide-to-ai-chatbot-development-tools-tips-and-best-practices-for-developers-a4768d8274a"> my previous article</a>. Please check it out in order to understand what “embedding”, “vector store”, “semantic search” is. If you prefer a video,<a href="https://www.youtube.com/watch?v=kMtf9sNIcao"> here</a> is something that will help.</p><h4>The process goes like this:</h4><ul><li>You have your documents and you split them into chunks.</li><li>You embed the chunks and you put the embeddings in a vector store.</li><li>You take the input from the user, embed it, and do a semantic search for relevant information in the vector store.</li><li>You give what you find + the user question to the LLM in order to formulate an answer.</li></ul><h4>How you will do it:</h4><ol><li>Install FlowiseAI onto your computer. Take a look<a href="https://docs.flowiseai.com/getting-started"> here</a> and see how you can do that.</li><li>After opening Flowise, just create a new chatflow for your AI Chatbot</li><li>Start off with a chain node (add it from the left “+” symbol, “Chains’’ category), more specifically a “Conversational Retrieval QA Chain”. This chain is used to query a knowledge base and it needs an LLM and a Vector store in order to work.</li><li>Add a vector store (choose “In-Memory Vector Store” if you just want to do some tests). The vector store will need the documents and an embedding function.</li><li>Add a document loader and upload files from your computer.</li><li>Attach a text splitter to the documents loader in order to split the text into smaller chunks that will be fed to the LLM.</li><li>Now that we have the documents, we need to embed the text so the AI can make sense of the information. Add an embedding function. Let’s say OpenAI Embeddings.</li><li>We’ve gone through most of the flow, now we just need to connect everything with an LLM. You can do that by going to “Chat Models’’ and selecting a model (let’s say ChatOpenAI).</li><li>Save the chatflow and press the message icon on the top left in order to test out the AI Chatbot.</li><li>Press on the “&lt;/&gt;” icon in order to see embedding options. Just choose the one that you are comfortable with. The easiest way to get the AI Chatbot on your website is to use the HTML code provided in the “Embed’’ tab. Just copy the code and insert it at the beginning of your &lt; body&gt; tag of your main HTML file.</li><li>You can also change the appearance of the chatbot by adding different attributes in the code. Check out<a href="https://www.youtube.com/watch?v=XOeCV1xyN48&amp;list=PL4HikwTaYE0HDOuXMm5sU6DH6_ZrHBLSJ&amp;index=7"> this video</a> with a full explanation of both step 10 and 11.</li></ol><p>And there you have it. You will need to go more in-depth with this tool if you want some more specific behavior. But, you should know that more complex flows can be created with this. Just take a look at<a href="https://www.youtube.com/watch?v=tD6fwQyUIJE&amp;list=PL4HikwTaYE0HDOuXMm5sU6DH6_ZrHBLSJ&amp;pp=iAQB"> this playlist</a> and go through the videos in order to build your AI Chatbot.</p><h3>WhatsApp AI Chatbot for small businesses</h3><p>You are a medical clinic with a small website full of information regarding your services, your doctors, prices, and many more. You receive a lot of questions on your WhatsApp number from your customers that could easily be answered if they just looked closer on the website. If that’s the case, why not use the information on the website in combination with an LLM in order to build an AI Chatbot that can answer your customers?</p><p>The problem is similar to the one in the previous example. But let’s use a different tool. We’ll go with<a href="https://www.voiceflow.com/"> Voiceflow</a> because you will be able to find a lot of tutorials for this.</p><h4>Here is how you do it:</h4><ol><li>Register to<a href="https://www.voiceflow.com/"> Voiceflow</a> and create a free account.</li><li>Start by creating an assistant using the “New Assistant” button on the top right corner of the page.</li><li>Choose “Build AI Assistant” and then pick WhatsApp when you are asked about channels and the English language.</li><li>Remove the predefined blocks but leave the starting node.</li><li>Go to “Knowledge Base” on the left menu and add the different pages of your website to the knowledge base by pressing “Add Data Source” → “URLs”. Then paste in all of the different URLs that you want, one by one.</li><li>Go back to the “Designer” space using the left menu and let’s start. First, connect a “Text” block (under the “Talk” category) to the “Start” block. Add whatever you want as a first message from the AI Chatbot to the user. You can also use a built-in feature to generate similar messages automatically and let the chatbot use a different one each time.</li><li>Add a capture card to the block from the tools menu, listed under “Listen” → “Capture”. Then add a variable called {userQuestion} in the right menu of the capture card. This will record the user’s question in a variable that we can use in other blocks.</li><li>From here, pass the user question into a new block. You want to send the question to the LLM so you need to choose the “Response AI” card from the “AI” category and create a new block with that card.</li><li>In this card you should select “Knowledge Base” as a Data Source and pass it the variable {userQuestion}.</li><li>Now you have the answer to the user’s question so let’s keep it going and ask the user if there are any other questions. Do this by adding a “Text” cell in the block followed by a “Choice” cell from the “Listen” category. Create 2 paths: yes and no.</li><li>If the user says no just add another “Text” card with a goodbye text like: “Have a great day!”. We want to be polite. You can also generate more ways of saying this so we get different messages.</li><li>Grab an “End” card from the “Logic” category and add it to the block. This will end the conversation.</li><li>If the user still has questions, just add another block with a “Text” cell asking “What can I do for you?”. Then “Capture” the answer and use the same variable as before, {userQuestion}. You just create the same block as in steps 6 and 7.</li><li>Now loop back to the “Response AI” card from step nr. 8 and that’s about it. Now the user can ask as many questions as needed and always get the answer from the AI Chatbot.</li><li>You can now test it by pressing the “Run” button on the top right of the page, selecting “Test on WhatsApp” and entering your phone number so you can chat with it in the WhatsApp app.</li><li>You can also publish it by following the steps of the “Publish” option.</li></ol><p>So, that’s about it. It’s not that hard to build such an AI Chatbot. It can be really useful if you are wasting a lot of time answering simple questions from your users.</p><h3>Other things you can do</h3><p>I’ve only shown you some very basic use-cases but these AI Chatbots can get really complex, really fast. For example, you can:</p><ul><li>Promote your Youtube channel. If the user has more than 5 questions regarding what you are selling or your services, just send a second message after the answer telling him “I see you are interested in the subject. Why not take a look at my Youtube channel where I share different aspects of what I do.”</li><li>Use tools like Zapier and connect them to a Google Spreadsheet. When the user gives his email to the chatbot (for example in order to receive a free ebook) you can take that answer and put it in a leads spreadsheet that you created.</li><li>Understand the intent of the user and react accordingly. If, for example, the user is asking how to return a product, you can create the flow of the conversation so that the AI Chatbot asks what is wrong, if the user is not satisfied, etc. It will gather specific information that it can send through a Zapier connection to another tool where you collect feedback.</li><li>Track orders by giving it access to your orders so that, when the user asks the status of order #1234, the AI Chatbot can answer with: “It left our storage today and is on its way to arriving at your address.” It can then handle follow up questions like “What address did I choose for delivery” or anything else the user might need to know.</li></ul><h3>Tools for AI Chatbot Development</h3><p>I’ve only talked about FlowiseAI and Voiceflow but there are many more tools that you can use. They all follow the same principles even though they have different flows and ways of helping you build chatbots.</p><p>Here is a list of some other tools you might want to check out:<a href="https://docs.langflow.org/"> LangFlow</a> (similar to FlowiseAI),<a href="https://botpress.com/"> botpress</a> (similar to Voiceflow),<a href="https://wotnot.io/"> WotNot</a> or<a href="https://n8n.io/"> n8n</a> (more of an automation tool).</p><p>At the moment these tools offer similar things so choose one that you like best. Start understanding how to map conversations and you will be able to switch between tools with no problem if needed.</p><p>I hope this helped you on your AI Chatbot development journey. Have a nice day and see you around!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ae07b26dbc16" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Beginner’s Guide to AI Chatbot Development: Tools, Tips, and Best Practices for developers.]]></title>
            <link>https://medium.com/@chirila.andr/beginners-guide-to-ai-chatbot-development-tools-tips-and-best-practices-for-developers-a4768d8274a?source=rss-5eeebaecb6f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/a4768d8274a</guid>
            <category><![CDATA[chatbots]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[chatbot-development]]></category>
            <category><![CDATA[ai-chat-bot]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Andrei Chirila]]></dc:creator>
            <pubDate>Fri, 29 Sep 2023 07:39:37 GMT</pubDate>
            <atom:updated>2023-09-29T07:39:37.383Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6H5y7xRtYR95qAsQ_BAbDg.gif" /></figure><p>So, you want to build an AI chatbot. Maybe it was my <a href="https://medium.com/@chirila.andr/a-chatbot-for-any-occasion-how-ai-chatbots-can-help-your-business-grow-a0ea89ea25b3">previous article</a> on chatbot use-cases that led you here, or you already had the idea in your head and just need a few tips regarding how you can do it. Well, you have come to the right place.</p><p>Before starting, just a quick note that this article is <strong>designed to help tech people</strong> (meaning people that are comfortable with coding) <strong>start an AI Chatbot Development project</strong>. I will assume you are comfortable with one programming language (Python or JavaScript), know what an API is, how to use one, and what a LLM is.</p><p>Mostly, I will be talking about 3 different ways to develop an AI Chatbot:</p><ul><li>using an API</li><li>fine-tuning a model</li><li>using RAG</li></ul><p>If you are looking for a way to <strong>build a chatbot that uses AI but aren’t comfortable with coding</strong>, then take a look at my other article meant for non-developers. (coming soon)</p><p>Now let’s get down to business:</p><h3>The “I want ChatGPT on my website” chatbot</h3><p>This is the easiest option. You practically just want your client to be able to talk to ChatGPT directly from your website. Maybe you have a pirate-themed website and just want to offer the user the possibility of talking to a “real live pirate”. If you ignore the possibility of hallucinations (it might invent dates or historical events), then this is the perfect option. Easy to implement, low costs, large knowledge base.</p><p><strong>Here is how you do it:</strong></p><ol><li>You build the interface of the AI chatbot using HTML, CSS, JavaScript, React or any other framework of your choosing.</li><li>Choose your LLM taking into consideration the capabilities and costs. Check them out on<a href="https://huggingface.co/"> HuggingFace</a>.</li><li>Get an API key. (<a href="https://huggingface.co/docs/hub/security-tokens">HuggingFace</a> or<a href="https://www.howtogeek.com/885918/how-to-get-an-openai-api-key/"> OpenAI</a>)</li><li>Check the documentation and see how you will need to send the information to the API. For example, for ChatGPT you will need to format it somewhat like in the code below.</li></ol><pre>let conversationHistory = []; <br>// store here all the user questions and the answers from GPT to have context of previous questions and answers<br>// store as { role: &quot;user&quot;, content: userMessage } for user message<br>// store as { role: &quot;assistant&quot;, content: gptMessage } for answers<br><br>const systemMessage={<br>    role:&#39;system&#39;,<br>    content:&#39;Answer me like a pirate&#39;<br>} <br>// this will be the initial prompt.<br>// it can be simple like in example or very complex with multiple instructions<br><br>const generateResponse = (chatElement) =&gt; {<br>    const API_URL = &quot;https://api.openai.com/v1/chat/completions&quot;;<br>    const messageElement = chatElement.querySelector(&quot;p&quot;); <br>    // or however you decide to extract the text from the user input<br>    <br>    conversationHistory.push({ role: &quot;user&quot;, content: userMessage });<br>    // push user message to conversationHistory before making the request<br><br>    // Define the properties and message for the API request<br>    const requestOptions = {<br>        method: &quot;POST&quot;,<br>        headers: {<br>            &quot;Content-Type&quot;: &quot;application/json&quot;,<br>            &quot;Authorization&quot;: `Bearer ${YOUR_API_KEY}`<br>        },<br>        body: JSON.stringify({<br>            model: &quot;gpt-3.5-turbo&quot;,<br>            messages: [systemMessage, ...conversationHistory],<br>        })<br>    }<br><br>    // Send POST request to API, get response and set the reponse as paragraph text<br>    fetch(API_URL, requestOptions)<br>        .then(res =&gt; res.json())<br>        .then(data =&gt; {<br>             messageElement.textContent = data.choices[0].message.content.trim();<br>             // although it looks weird .choices[0],message.content is where you will find the text answer to the request<br>             conversationHistory.push({ role: &quot;assistant&quot;, content: messageElement.textContent });<br>             // add the answer to the conversation history<br>    }).catch(() =&gt; {<br>        messageElement.classList.add(&quot;error&quot;);<br>        messageElement.textContent = &quot;Oops! Something went wrong. Please try again.&quot;<br>}</pre><p>There might be better ways to implement this. I just wanted to give you a quick glimpse of how to send and receive messages to and from the API in order to build your AI chatbot. Now just show the response to the user and you are done!</p><p>For a more detailed explanation of how to work with OpenAI API,<a href="https://platform.openai.com/docs/guides/gpt"> check out the documentation.</a></p><h3>The “Still ChatGPT but with more specific answers” chatbot</h3><p>If you can’t get the ChatBot to behave the way you like through prompting, you might want to think about fine-tuning a model. Fine-tuning is using a pre-trained model and training it on a dataset specific to your task. You will obtain an AI chatbot with more specific answers to your use-case. However, it will cost you more time and money.</p><p>So, before fine-tuning, try to find other ways of solving the problem. Here are some<a href="https://platform.openai.com/docs/guides/gpt-best-practices"> strategies for getting better results with ChatGPT</a> that you can try before fine-tuning. If prompt engineering or prompt chaining (breaking complex tasks into multiple prompts) doesn’t work, then the next step is fine-tuning.</p><h4>Usecases for fine-tunning an AI Chatbot can be:</h4><ul><li>Setting the style and tone of the AI Chatbot <em>(you want to make it sound like a sarcastic pirate but prompting just doesn’t give you the specific type of sarcasm that you want)</em></li><li>Getting it to answer in a specific format <em>(maybe you want to feed it text and the chatbot will answer only with bullet points containing specific information from that text like persons mentioned in the text, tasks mentioned, deadlines etc.)</em></li><li>Getting the AI chatbot to follow complex prompts <em>(this can be useful if you want to increase security, reduce hallucinations and jailbreak possibilities)</em></li><li>Handling edge cases in specific ways <em>(you can specify exactly how you want it to answer in certain situations)</em></li><li>Performing tasks that are hard to articulate in a prompt</li></ul><p>Depending on the LLM that you want to finetune, the process will vary. You can use TensorFlow with Keras, PyTorch or just Python.<a href="https://huggingface.co/docs/transformers/training"> Here</a> you can find a <strong>tutorial for fine-tuning on HuggingFace</strong>, and<a href="https://platform.openai.com/docs/guides/fine-tuning"> here</a> is the <strong>documentation for OpenAI fine-tuning</strong>.</p><p>No matter what LLM you choose, you will need 1 thing to fine-tune your AI chatbot: <strong>a dataset.</strong></p><ul><li>the dataset you will use will determine the way the model is fine-tuned and the manner in which your AI chatbot will answer questions.</li><li>the dataset should be in the style of Question — Answer</li><li>you should have <strong>at least 50–100 entries</strong> in the dataset in order for the model to be trained efficiently</li><li>make sure you include all relevant questions that you think your customers might ask, all edge-cases that you want to be treated differently and all questions that might pose security risks.</li><li>the model will learn from your dataset how to answer specific questions but also in general (style, tone) and will offer a better result than the normal model on specific queries from the user.</li></ul><p>If you want to be extra specific, you can always include prompts when using the model in your AI chatbot, just like you do when you use the regular model. After fine-tuning, you will receive a model name which you can just insert in your API request like this</p><pre>const systemMessage={<br>    role:&#39;system&#39;,<br>    content:&#39;Answer me like a pirate&#39;<br>} <br>// prompt it however you like<br>// use prompt engineering techniques on top of fine-tuning for better results<br><br>const requestOptions = {<br>        method: &quot;POST&quot;,<br>        headers: {<br>            &quot;Content-Type&quot;: &quot;application/json&quot;,<br>            &quot;Authorization&quot;: `Bearer ${YOUR_API_KEY}`<br>        },<br>        body: JSON.stringify({<br>            model: &#39;${YOUR_MODEL_NAME}&#39;,<br>            messages: [systemMessage, ...conversationHistory],<br>        })<br>    }</pre><p>If you want to fine-tune an OpenAI model <strong>I also recommend using this</strong><a href="https://colab.research.google.com/drive/16e_HJwCHNsATFVtljZTMg0JydrHAhO8A?usp=sharing"> <strong>Google Colab project</strong></a> that takes you step by step through everything you need to do and even creates the dataset for you.</p><h3>The “ChatGPT but he knows my business” chatbot</h3><p>The issue with normal models and even fine-tuned models <em>(although you can add some business-specific information when fine-tuning)</em> is that they run on a general knowledgebase. This can be useful for the way you want your AI Chatbot to behave, or it can completely suck and make it useless.</p><p>For example, let’s think of the following scenario. You have an online store with hundreds of products and want to create an AI Chatbot for your customers. You won’t be able to put all of your information about your products in a prompt and would need thousands of entries in your dataset for fine-tuning. The solution is simple: “Give up! This is not for you.”</p><p>Just kidding. There is a way to train a model specifically on your knowledgebase (that can be thousands of documents) and create an AI Chatbot specific to your business.</p><p><strong>Retrieval-augmented generation (RAG)</strong> is an AI framework for improving the quality of LLM-generated responses by grounding the model on external sources of knowledge. The RAG Chatbot development is more difficult, but not so much that it is something you need to be scared of.</p><p>There are multiple ways of building a RAG Chatbot but the main steps are the same:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s_byPkTUoATNfNSnPfJumQ.png" /></figure><ul><li>The first step is to split all of the information into manageable chunks of text. You will send the most relevant chunks alongside the user question to the LLM. That is why you should use chunks around 1000–2000 tokens.</li><li>Secondly, LLMs don’t really use text but<a href="https://huggingface.co/blog/getting-started-with-embeddings"> embeddings</a> which are a vectorised representation of information (text in our case). You will need to run all of the chunks through an Embedding Model in order to vectorise it. Take into consideration that the OpenAI embedding model is only ranked 13 (at the time of writing this) on the<a href="https://huggingface.co/spaces/mteb/leaderboard"> Embedding Leaderboard</a>. Take a look at<a href="https://huggingface.co/blog/mteb"> how they are ranked</a> and choose the model that best suits you.</li><li>You will then need to add the embeddings into a Vector Store of your choosing.<a href="https://www.pinecone.io/"> Pinecone</a> is the most popular but there are open source alternatives like<a href="https://docs.trychroma.com/"> Chroma</a> or<a href="https://qdrant.tech/"> Qdrant</a> that you can use. Take a look at<a href="https://objectbox.io/vector-database/"> this article</a> to better understand what a Vector Store is and which you should use.</li><li>The same needs to be done with the user’s question — run it through an embedding model</li><li>You will then perform a semantic search on the Vector Store in order to receive the text chunks that contain the most relevant information, ranked by their relevance to your question.</li><li>The question and the relevant chunks of text are then fed into the LLM of your choosing which can do its magic and create a nice answer to your question.</li><li>The LLM can also be prompted in order to answer in specific ways and even fine-tuned if you wish.</li></ul><p>You can do this in many different ways using Python or JavaScript. I recommend you use<a href="https://www.langchain.com/"> LangChain</a> framework or<a href="https://embedchain.ai/"> EmbedChain</a> as they make the process a lot easier.</p><p>Here is an example using Python, LangChain and streamlit with FAISS as a vector database:</p><pre>from dotenv import load_dotenv<br>from PyPDF2 import PdfReader<br>from langchain.text_splitter import CharacterTextSplitter<br>from langchain.embeddings.openai import OpenAIEmbeddings<br>from langchain.vectorstores import FAISS<br>from langchain.chains.question_answering import load_qa_chain<br>from langchain.llms import OpenAI<br>import streamlit as st<br><br>def main():<br>    load_dotenv()<br>    st.set_page_config(page_title=&quot;Chat with a PDF&quot;)<br>    st.header(&quot;Chat with a PDF 💬&quot;)<br>    <br>    # upload file that you want to &quot;talk&quot; to<br>    pdf = st.file_uploader(&quot;Upload your PDF file here&quot;, type=&quot;pdf&quot;)<br>    <br>    # extract the text and put it in a variable<br>    if pdf is not None:<br>      pdf_reader = PdfReader(pdf)<br>      text = &quot;&quot;<br>      for page in pdf_reader.pages:<br>        text += page.extract_text()<br>        <br>      # split into chunks with chunk size 1000 and overlap 200<br>      char_text_splitter = CharacterTextSplitter(separator=&quot;\n&quot;, chunk_size=1000,<br>                                                 chunk_overlap=200,length_function=len)<br>      text_chunks = char_text_splitter.split_text(text)<br>      <br>      # create embeddings and upload them into FAISS vector database<br>      embeddings = OpenAIEmbeddings()<br>      docsearch = FAISS.from_texts(text_chunks, embeddings) <br>      llm = OpenAI() <br>      chain = load_qa_chain(llm, chain_type=&quot;stuff&quot;)<br>      <br>      # take user input and make similarity search in vector database<br>      query = st.text_input(&quot;Type your question:&quot;)<br>      if query:<br>        docs = docsearch.similarity_search(query)<br>        response = chain.run(input_documents=docs, question=query)<br>           <br>        st.write(response)<br><br>if __name__ == &#39;__main__&#39;:<br>    main()</pre><h3>Benefits for each implementation</h3><p>It’s time to put all 3 of the possible AI Chatbot implementations and see what the pros and cons are for each one.</p><h4>Costs</h4><ul><li><strong>API call:</strong> the costs for this implementation are the lowest as you only pay for the number of tokens used in questions &amp; answers.</li><li><strong>Fine-tune:</strong> the costs for fine-tuning are higher. You not only pay in order to fine-tune the model but, in the case of OpenAI, you pay around 4x more for tokens used in questions &amp; answers for fine-tuned models.</li><li><strong>RAG:</strong> the cost here is also bigger than the simple API call and might also be bigger than fine-tuning. It will depend on the amount of information in your knowledge base. You will need to pay for embedding the information and also the number of tokens that you feed the LLM with when searching for an answer to the question.</li></ul><h4><strong>Development difficulty:</strong></h4><ul><li><strong>API call:</strong> very easy to implement. Beginner level.</li><li><strong>Fine-tune</strong>: easy to implement but might require a lot of work in order to create the dataset.</li><li><strong>RAG:</strong> medium-level difficulty, will need some experience with Python or LangChain framework.</li></ul><h4><strong>Interaction with the user</strong></h4><ul><li><strong>API call:</strong> will interact just like ChatGPT (or any other model) when prompted to answer in a specific way. It will offer mostly general information.</li><li><strong>Fine-tune:</strong> more custom interaction. Besides classical prompting, it can handle edge-cases better. It can be fine-tuned in order to be more secure and harder to jailbreak or hallucinate. It can also answer specific business questions if they are present in the dataset but will mostly offer general information. This will depend on how big and complex the dataset you fine-tune it with is.</li><li><strong>RAG:</strong> best for situations where you want precise answers with information from a custom knowledge base. This will offer the best user interaction if the user wants to find out detailed information about your business and not just general information.</li></ul><p>_________________________________________________________________</p><p>So, this is it. These are the 3 techniques I would use for AI Chatbot development, if I would like to get my hands dirty and write some code. There are no-code (or low code) alternatives that I will be presenting in a future article (coming soon). If you want some inspiration regarding use-cases for chatbots, then<a href="https://medium.com/@chirila.andr/a-chatbot-for-any-occasion-how-ai-chatbots-can-help-your-business-grow-a0ea89ea25b3"> take a look at this article</a> that is already published.</p><p>Have a nice day!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4768d8274a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Chatbot For Any Occasion: How AI Chatbots Can Help Your Business Grow]]></title>
            <link>https://medium.com/@chirila.andr/a-chatbot-for-any-occasion-how-ai-chatbots-can-help-your-business-grow-a0ea89ea25b3?source=rss-5eeebaecb6f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0ea89ea25b3</guid>
            <category><![CDATA[chatbot-development]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[chatbots-for-business]]></category>
            <category><![CDATA[ai-chatbot]]></category>
            <category><![CDATA[chatbots]]></category>
            <dc:creator><![CDATA[Andrei Chirila]]></dc:creator>
            <pubDate>Mon, 18 Sep 2023 10:53:59 GMT</pubDate>
            <atom:updated>2023-09-29T11:21:59.272Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rp1sa7uSSCbg2kMVU004jQ.png" /></figure><p>One of the biggest winners of the recent AI advancements are, without a doubt, chatbots. I mean, even <a href="https://openai.com/blog/chatgpt">ChatGPT</a>, the tool that started a general months-long worldwide hype, could be described as a simple chatbot with a lot of knowledge.</p><h3>Why are chatbots such a big winner?</h3><p>It’s easy. One of the biggest advancements in the last year has come in the shape of LLMs (large language models). And, I mean, it’s in the name. Computers have learned how to understand language and how to summarize and create text (which is, practically, language). So any use case that implies using language (text mostly, although voice cloning isn’t far behind) has received a couple of new AI tools that are on par with what a human can do. Pair this with integrations that connect different tools (through Zapier for example) and all of a suddent you have a helpful assistant that can take questions, look for information, offer short and precise answers in a matter of seconds, send the information to where it needs to get and keep you updated with what’s been happening while you were away.</p><h4>But people won’t like it! There is nothing like human communication.</h4><p>Wrong! Kipp Bodnar (HubSpot’s CMO) <a href="https://youtu.be/htKMqgVgKvc?t=664">said</a> that they managed to implement a chatbot that was able to <strong>answer 78% of users’ questions</strong>, which, in turn, led to an <strong>almost 300% (!!) increase in client satisfaction</strong>. You do the math on what this could mean for your business.</p><p>Most of the users’ questions fall in the category of simple, procedure related or information seeking questions (“How do I do X?” “Is Y feature available with your product?”). An answer from a chatbot that is always friendly and can offer the correct information in a couple of seconds instead of waiting minutes or hours for a human to answer is, most of the time, enough for the user to be satisfied. And don’t underestimate the WOW factor when they realize the level of complexity the chatbot can handle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*2jwwz0r0PAmYz0hn2GaP9w.gif" /></figure><h3>There is more than 1 type of chatbot?</h3><p>Well, yes. When you think of a chatbot you probably think of a small round widget in the bottom right part of the screen that appears on some websites. And you would be right for 80% of cases. But the complexity of these AI chatbots has increased and they can take different forms now.</p><p>Chatbot development platforms will let you integrate them with WhatsApp, Telegram or even Facebook Messenger. You can also integrate it with your email and let it answer questions sent to the <a href="/service/mailto:support@your-busines.com">support@your-busines.com</a> email address. They can take information from the conversations they are having and fill in spreadsheets with data for lead generation. Also, you can easily find out what are the most common problems your users are facing or their general feeling towards your product.</p><h4>Support chatbots.</h4><p>This is the most obvious type of AI chatbot. The same round widget on the bottom right corner of your page that the visitors of your website can access and get answers from. The difference? An AI chatbot is far more versatile than the “old generation” one. It can answer a bigger range of questions. Also, it can offer different types of support (like offering the user step-by-step instructions, a link to what he looking for, or even a video explaining how to do certain things) depending on what the user needs:</p><ul><li>It knows your business.</li><li>The chatbot has infinite patience with your customers.</li><li>It can explain the same thing in 10 different ways.</li><li>Answering questions takes only a couple of seconds.</li><li>It’s available 24/7.</li></ul><p>Different chatbot development tools will help you with a wide variety of actions that the chatbot can do. For example, it can ask for emails or telephone numbers if it feels that the user is interested in buying. Then, it can add them to a predefined Google Spreadsheet creating a list of possible clients that you can contact.</p><p>Or, if it feels that the customer isn’t satisfied with the order or the delivery time, it can offer a 10% discount for a future order. The chatbot can do this by generating a discount code if it’s integrated with a tool that can generate these codes.</p><h4>Onboarding chatbots.</h4><p>Maybe you have a big business with hundreds of employees. Congratulations, by the way! Onboarding each new employee might take a lot of time. Besides that, you probably have a lot of documents with procedures for each type of activity that an employee needs to do. It will be hard for them to remember everything from the first day. They will have questions. That means more time wasted answering these questions which sometimes look like: “You need to look in doc. 12 I think, page 15 or 25 and you will find it”.</p><p>Well, a chatbot can answer all of these questions. Additionally, it will provide precise information (doc. 12 page 35, not 15 or 25, besides the fact that it will do it in a couple of seconds). Moreover, it will be able to give a direct text summarisation of the procedure so the employee avoids wasting time looking for the document, downloading, finding the correct page, etc.</p><p>But onboarding isn’t only for new employees. It’s for new customers also. Your platform might be complicated, offering a lot of services to your customers. It might be difficult for them to find what they are looking for. Or maybe they aren’t sure how to do a certain thing or if it’s possible. An AI chatbot can help with that and increase customer retention by guiding them step by step in order to set up their account and customize it exactly to their needs. Your platform suddenly isn’t that complicated anymore.</p><h4>Process improvement.</h4><p>Yes, you can use a chatbot not only to talk to it and get fast answers to your questions, but also to improve some of your processes. Let’s say that you need to constantly update multiple spreadsheets after doing certain tasks. You probably need to go to your laptop so you can easily find the links, open the spreadsheets, remember the exact information, fill in the correct cells, etc.</p><p>But what if the chatbot would do it for you while you just type the information in a WhatsApp conversation? So you just do the task, you take out your phone, and you write the details in a straightforward format. For example: “At 10:45, me and Matt finished installing the network cables on the 4th floor of the Main Office building. There were no issues.”. Afterward, the chatbot will send the information to a spreadsheet where the hour, person responsible, task, location, and comments sections will be automatically filled in. You can even take a picture and it will be attached to your comments.</p><h3>But what are the use-cases?</h3><p>Are you still not convinced? I respect that. I was a little bit of a doubter myself until I started looking at different types of businesses and understood that ANY TYPE OF BUSINESS can find a way to use a chatbot. Let me give you some examples from the top of my head:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*MemTALkptxjdVa5Zni3lSQ.gif" /></figure><h4>Airbnb hosts</h4><p>Just teach the chatbot what the most asked questions are, integrate it with WhatsApp, and leave the number on your Airbnb page for clients to see. Voilà, you just got rid of 80% of the questions you had to answer, like:</p><ul><li>“How do I get there from the airport?”</li><li>“Do I need to bring my own towels?”</li><li>“Could you recommend a restaurant nearby?”</li><li>“Is it the white door or the yellow one?”</li></ul><p>If there is any question that can’t be answered by the chatbot, the clients can text you instead. You can leave your phone on silent and not worry anymore that your “communication” rating will go down while your clients get fast answers to their questions. Win-win situation. Multiply this by the number of apartments you have and all of a sudden you have a lot of free time on your hands.</p><h4>Recruitment</h4><p>You often have open positions available at your company and receive a lot of CVs, but you notice that most of the candidates aren’t a fit for any of your positions. What if they go through a short talk with an AI Chatbot that can help them understand if any of the positions available match their skills? <strong>No selection!</strong> The candidates can send their CV even if they don’t talk with the chatbot. Still, it’s important that they have the support of a friendly figure who tries to explain exactly what each of the job openings is for and what skills would be needed in order to get that job. Or, the other way around, candidates can tell the chatbot their skills (or maybe they can upload their CV directly) and the chatbot can answer with job openings that might suit them.</p><h4>Safety and Security</h4><p>Safety and Security. Let’s take the example of a construction company where you often face safety risks. One of your employees notices that on the second floor of the building he is working on there is a crack in the wall due to drilling on the other side. He doesn’t have the expertise or the time to fix it, and doesn’t know who has. But he has WhatsApp installed on his phone and can just log in the issue using the WhatsApp bot that transfers it to a list where the person responsible can easily see it, see a picture of it, when it was added, who added it, if it’s urgent or not, and make a decision regarding fixing it.</p><h4>Restaurants and bars</h4><p>Just give the chatbot access to your menu with details regarding all of your dishes and beverages. Then, let the customers have profound conversations regarding what they want to order: “I am on a diet and I want a main course and a dessert that have less than 800 calories. I’m also lactose intolerant. I would also like a wine that pairs well with the main dish.” They can do this from home, on their way to the restaurant, or while sitting at the table.</p><h4>Education</h4><p>This one is easy. Your business offers knowledge in one way or another. When people enter your website they want to learn and find answers to their questions. Do I need to say more? A chatbot can help them navigate through all of the information that is available on your website, can help them make study plans and show them the benefits of buying one of your plans.</p><h3>But how do I build an AI chatbot?</h3><p>It’s easy, really easy. But I can feel that you are getting tired after reading all of this text. So I will give you a break. You can continue reading <a href="https://medium.com/@chirila.andr/beginners-guide-to-ai-chatbot-development-tools-tips-and-best-practices-for-developers-a4768d8274a">my next article</a> if you are a tech person and want to know the tech details of developing an AI chatbot. And if you are a non-tech person and want to know about no-code chatbot development tools, you can continue with the third article (soon to come).</p><p>So take a break and continue with one of the 2 articles that sounds the best for you.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0ea89ea25b3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The untapped potential of chess (and how it’s being tapped into)]]></title>
            <link>https://medium.com/@chirila.andr/the-untapped-potential-of-chess-and-how-its-being-tapped-into-6dd4fe237d4b?source=rss-5eeebaecb6f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/6dd4fe237d4b</guid>
            <category><![CDATA[chess]]></category>
            <category><![CDATA[games]]></category>
            <category><![CDATA[online-games]]></category>
            <category><![CDATA[esport]]></category>
            <category><![CDATA[board-games]]></category>
            <dc:creator><![CDATA[Andrei Chirila]]></dc:creator>
            <pubDate>Mon, 07 Sep 2020 11:21:47 GMT</pubDate>
            <atom:updated>2020-09-09T09:30:09.656Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*T9gxidmdykAgHI23D1cJxg.png" /></figure><p>I learned to play chess when I was 6 years old and, although I was never a great player, I enjoyed the occasional game. The problem was that, as a child, it was hard for me to find someone that was my age with whom I could practice. I have this image in my mind of a small park next to the building that I lived in, where old men (I don’t remember seeing a woman ever) would meet and play chess for hours. They would be gathered around a table in silence, watching whoever was playing and that’s how I started viewing the stereotypical chess player as an old, slow, constantly thinking man.</p><p>For a couple of months now I’ve started being interested in chess again. It’s something that comes and goes — last time this happened was about 4–5 years ago — but this time it’s different as I’ve found the chess world and the chess community in a whole different place than when I left it.</p><h3>Popularity going up</h3><p>Chess as we know it is changing.</p><p>Since the Covid-19 epidemic started everyone had to find ways to adapt to the new status quo. That meant that even the group of senior chess players that used to gather in the park next to where I lived had to change their daily routine and move online. Platforms such as chess.com or chess24.com saw a massive growth of their player base over the last couple of months because of this. But that’s not all that happened.</p><p>Besides old fans moving online, new fans are joining the hype. Twitch streamers like xQcow (3.3M followers) have started playing chess and taking lessons from famous players like GMHikaru, 5-time United States Chess Champion, #1 blitz player and one of the most known chess streamers. These collaborations are transforming chess streamers and youtubers into popular content creators giving the game a whole new look and feel — fun, exciting, challenging. Online, the chess community is becoming a lot more vibrant, friendlier towards new players and open to having fun and promoting the game. You can see GMs (Grand Masters) &amp; IMs (International Masters) playing together with their followers, commentating each others games or making instructional videos on openings or endgame tactics.</p><p>Already we’re starting seeing collaborations between different players that start showing up constantly on each other’s channels (like Hikaru and GothamChess for example) doing sub battles or playing all sorts of different variations and even doing “<a href="https://www.youtube.com/watch?v=pyM5MImsJVk">best of</a>” moments for laughs.</p><h3>Pogchamps 1&amp;2:</h3><p>16 Twitch streamers (x2 as there were 2 events) were invited to play in an online chess championship with a $50,000 prize fund, where the idea was that top rated chess players — Hikaru Nakamura, Daniel Rensch, Alexandra Botez, Qiyu Zhou, Daniel Naroditsky, Levy Rozman, Robert Hess etc.— would coach new players and commentate on their games.</p><p>This made for some very entertaining games as well known variety (xQc, Cr1tikal, Papaplatte), League of Legends (VoyBoy, Yassuo, Gripex90), Fortnite (Nate Hill, Swiftor), Heartstone (Forsen, itshafu) and even poker (easywithaces), politics (David Packman) and .. well he classifies as variety but best known as The Mountain / World’s Strongest Man (Hafþór Júlíus Björnsson) streamers went head to head with only a couple of months of preparation into this complex game. All of this made for some very popular Twitch and Youtube content as these new players got lessons in chess from some of the best players around, while the chess players got a lesson in entertainment and streaming from these well known content creators.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/910/1*g2LqDMZCfBvGENlB2pDr_g.png" /></figure><p>This collaboration helped both sides and here are some examples: the chess.com stream that streamed the event had around 100K viewers for the finals while xQc’s loss against Cr1tikal gathered over <a href="https://www.youtube.com/watch?v=e91M0XLX7Jw">3M views on Youtube</a>. Besides that, chess channels such as <a href="https://www.youtube.com/user/wwwChesscom">Chess.com</a>, <a href="https://www.youtube.com/channel/UCweCc7bSMX5J4jEH7HFImng">GMHikaru</a>, <a href="https://www.youtube.com/channel/UCAn8NrZ-J4CRfwodajqFYoQ">BotezLive</a>, <a href="https://www.youtube.com/channel/UCQHX6ViZmPsWiYSFAyS0a3Q">GothamChess</a>, and even chess channels of people not directly involved in Pogchamps like <a href="https://www.youtube.com/user/AGADMATOR">agadmator</a>, <a href="https://www.youtube.com/user/RosenChess">Eric Rosen</a>, <a href="https://www.youtube.com/channel/UCAov2BBv1ZJav0c_yHEciAw">Samay Raina</a> and <a href="https://www.youtube.com/user/themagnuscarlsen">Magnus Carlsen</a> (nr. 1 chess player) have seen a dramatic increase in subscribers and viewers in June (first Pogchapms tournament) and August (second Pogchamps tournament). Streamers also benefited from this. Ludwing, for example, became for a brief moment the biggest chess streamer with a total of 71K concurrent viewers — when he usually has a maximum of 10K.</p><p>All in all, these events brought chess to a younger fanbase and helped chess channles with educational and entertainment videos grow. The collaboration between popular streamers and chess professionals transformed the game and revitalised it’s fan base while making the games more exciting. That’s why, from a niche, chess looks like it’s becoming a popular game again.</p><h3>And there is more</h3><p>Up until now we’ve only talked abut popularity. But there is antorher type of change coming.</p><p>A couple of weeks ago a big announcement was made. GH Hikaru has signed with the well known esports team TSM. This is a big move as chess has now entered the esports arena which was worth over USD 1072.35 Million in the year 2019 and keeps on growing. After Hikaru joining TSM, news of other popular chess players having talks with esports teams have started showing up and were confirmed after Qiyu Zhou (<a href="https://www.twitch.tv/akanemsko">akaNemsko</a>), signed with CLG.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/696/1*gurfB1vcl5rDzg4oNwXPPA.jpeg" /></figure><p>And here is where I think change will start coming for chess. Esports has been growing in the last couple of years because it knows how to keep people entertained. A lot of people with experience in gaming, streaming and content creation are part of the esports world and it seems that they have their eyes set on chess.</p><p>Game modes such as 4 player chess or 5D chess have already started becoming more popular as streamers and youtubers have found an audience interested in these less serious / more for laughs types of chess. The entertainment aspect seems to be growing as chess players are trying to keep their newly found (and definitely younger) audience hooked on chess content.</p><p>But this in not a direct impact of chess becoming an esport. What will probably be a direct impact is the way that chess events and competitions will look like from now on.</p><h3>Competitions</h3><p>Chess suffers from poor organisation when it comes to competitions and events. Just listen to what David Llada, FIDE’s Chief Marketing and Communication has to say about this:</p><blockquote>You give 10k to the average chess organizer to improve his tournament, and he will spend it on hiring two more 2600 players — like if this is going to make a difference. You give him a million, and he will organize ten tournaments in exactly the same way, with 100k in prizes each.</blockquote><p>And moving online only made things worst (at the moment). The 2020 Chess Olympiad had 2 instances of games ending due to connection issues and, for a part of the community, neither of them were resolved in a satisfying manner. This is even more problematic as one of the issues happened during the finals between India and Russia, FIDE deciding that both teams will be declared winners and will receive the 1st place.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/543/1*ywtWWpPwep0FnS4KX0Y_jg.png" /></figure><p>Poor organisation isn’t the only thing that chess tournaments struggle with. When it comes to prizes, compared to other sports, chess looks like the poorer cousin. He’s smart, he’s determined, but he just can’t find a way to make money and move out from his parent’s basement. This is where esports will probably have the biggest say as their experience with gaming will help popularise chess and, thus, bring more money into this sport.</p><p>At the moment this subject hasn’t been touched yet and remains a little bit unclear. There are a lot of jokes regarding what Hikaru joining TSM means. He is being taunted, in a friendly manner, that from now on people will be chanting ”TSM! TSM!” at his games (which usually happens for other esport games).</p><p>Leaving this asside, when it comes to competitions esports know how to make them big. Just look at the <a href="https://nexus.leagueoflegends.com/en-us/2019/12/2019-world-championship-hits-record-viewership/">League of Legends World Finals 2019</a> where there were 44 Million viewers at the peak of the event. Chess isn’t there yet but I imagine that, in the future, esport teams will want to break from the traditional chess competition format and invent their own style. I can imagine teams of chess players going head to head (similar to the Chess Olympiad), fighting to obtain the title for their organisation. This might mean that we will get to see chess rivals becoming teammates and who knows what else.</p><h3>A couple of obstacles</h3><p>There are a couple of roadblocks along the way and it’s not clear in chess will ever make it to the point of popularity and well organised competitions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*XRJpc5-NfHA-R4TblwW_Yg.png" /></figure><h4>Elitism.</h4><p>Probably the biggest problem that chess has is with it’s current player base and elitism. Chess is viewed as a game of “intelligence, sophistication, strategy, perfection and complexity” and, according to a <a href="https://www.chess.com/news/view/how-popular-is-chess-8306">study</a> done earlier this year, the people playing chess are well educated and value chess as a strategical game that is about “Thinking and Winning”, not entertainment. That makes the community a little bit elitist, not wanting to accept just anyone into their ranks as they might “spoil the game”. Pogchamps has found it’s fair share of doubters and haters for this reason alone and the new influx of players are definitely being watched from above by the “old guard”. On this subject it’s interesting to see <a href="https://www.youtube.com/channel/UCq6VFHwMzcMXbuKyG7SQYIg">penguinz0</a>’s video called <a href="https://www.youtube.com/watch?v=w_5RqOMwMRw&amp;ab_channel=penguinz0">Chess Drama</a>, which is a view on the problem from a new player perspective and, if you are still interested in the subject, <a href="https://www.youtube.com/channel/UCL5YbN5WLFD8dLIegT5QAbA">agadmator</a>’s response <a href="https://www.youtube.com/watch?v=ZE8XbnwfJDE">video</a> on this topic.</p><h4>Chess is a hard game</h4><p>While this might make it challenging for some, others will find it boring after a couple of games as it involves a lot of thinking and not that much action. Chess is a game that is easy to play but hard to master and, especially in this day of age when everyone argues that the attention span of people has become extremely short, not many people might have the patience to master the game.</p><h4>The chess establishment</h4><p>I also see a small problem coming from the institutions of chess. As this game has been around for centuries, there are many institutions that exist, governing the “laws” of chess. FIDE is by far the most important one and the role it will play in the future is essential. Will it support and work together with the players, youtubers, streamers and esports teams in order to make chess popular again and give it a new, fresh and exciting look, or will it try to hold on to what has been built up until now?</p><p>If these roadblocks will persist chess might remain at it’s current state: an unpopular game that everyone knows how to play. Esport teams might loose interest and youtubers and streamers will see their momentum disappear along with their audience.</p><h3>Check or Checkmate?</h3><p>Chess has a lot of momentum right now. Well known streamers like xQc, Ludwig and Cr1tikal have picked up the game and are sharing it with their audiences who, in turn, are learning more about the game and have started playing it. Chess content creators are also on the rise. While in the past only a couple of chess channels like agadmator, GMHikaru, Magnus Carlsen and BotezLive were managing to gather an audience, smaller channels like GothamChess, Samay Raina and Eric Rosen are seeing more and more popularity and even top players like <a href="https://www.youtube.com/channel/UC9Zs0DGUuenE23tObR50jJQ">Anish Giri</a> are joining the content creating wave. Esports has joined the arena and the promise of chess as a mainstream game is here.</p><p>What do you think? Are we going to hear about chess more often from now on? Are youtubers, streamers and esports going to make an impact or is it just a phase that we are going through? Is chess’ elitism going to scare away all the new players brought in by pogchamps, or are the haters going to hate while everyone else just enjoys playing the game?</p><p>For me, I’m just riding the wave right now and enjoying the content!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6dd4fe237d4b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Working with technical debt (a PM perspective)]]></title>
            <link>https://blog.devgenius.io/working-with-technical-debt-a-pm-perspective-679341f45065?source=rss-5eeebaecb6f4------2</link>
            <guid isPermaLink="false">https://medium.com/p/679341f45065</guid>
            <category><![CDATA[technical-debt]]></category>
            <category><![CDATA[project-management]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[risk-management]]></category>
            <dc:creator><![CDATA[Andrei Chirila]]></dc:creator>
            <pubDate>Mon, 22 Jun 2020 10:57:52 GMT</pubDate>
            <atom:updated>2020-06-24T08:00:17.316Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/1*OcnUZ2GRa3yPf6FfARMt6w.jpeg" /></figure><h3>What’s happening?</h3><p>Things haven’t been going so well lately.</p><p>It’s been a couple of months since you and your team were able to complete a sprint successfully, without leaving any tickets behind. Simple functionalities or minor changes seem to be taking way longer than they should and everyone seems to be underestimating how long it will take them to do things. The teams velocity keeps going down even though everyone seems to be working extra time to meet the sprint goal. Besides that, bugs keep showing up out of nowhere and the more you fix, the more pop up out of nowhere. Making any kind of schedule seems impossible and if it keeps going like this you can definitely say that you’re not going to hit your end of project deadline.</p><p>There might be a lot of reasons for projects not meeting their deadlines but if the situation sounds a little bit like the one above, you might be facing a case of : <strong>high technical debt.</strong></p><h3>Why is technical debt so scary?</h3><p>Well, it’s in the name: <strong>it’s debt</strong> and if it accumulates you might be <em>spending all the resources that you have into paying that debt rather than generating business value</em>.</p><p>I’ll spare you the technicalities. What’s important to know is that <strong>technical debt is normal on any project and it’s even useful</strong>. Whenever you need fast results and you “take a shortcut”, you create a little bit of debt. Sometimes, you will want to get fast results to meet a deadline, get a functionality out so it can be tested by users or just to show the stakeholders that you get things done and build some trust. That’s not a problem as long as you pay that debt (rewrite the code) sooner rather than later. Fast results now means more work later so be mindful about when you choose to go fast, and when you take time to fix things up.</p><p>But<strong> how do you know that you have a problem with technical debt</strong> and that you should start worrying about it? Well, I’ve already given a hint on how such a project looks like in my first paragraph, let me just put some bullet points in front so it’s easier to follow:</p><ul><li>functionalities or changes that seem easy on a first glance end up<em> </em><strong>taking a lot more time than expected</strong>;</li><li><strong>velocity</strong> is smaller than usual or is all over the place (some sprints it’s low, other sprints it’s back to normal) although not much changes from sprint to sprint;</li><li><strong>releases are getting buggier</strong> and the more time you are spending on bugs (anything over 20% should be a cause for concern) the more bugs seem to appear;</li><li>your tester seems to be overwhelmed at times and suddenly <strong>everyone starts doing testing</strong> to make sure all bugs are “caught” before the release;</li><li>sometimes, it seems that you are <strong>spending more time fixing stuff, rather than developing</strong>.</li><li><strong>estimates</strong> (even simple ones) are hard to get right even when you add a buffer and take into consideration the worst case scenario;</li><li>even though people seem to be working more than usual <strong>deadlines</strong> either aren’t met or, when they are, it is because of some superhuman effort;</li></ul><p>If most of these apply, then you probably have a problem that <strong>you need to fix. </strong>If only some of these apply or they apply to a lesser extent (for example not all releases are buggy but one in 2 or 3 are) then maybe you are somewhere on the edge and <strong>you need to make a decision</strong> — pay now or pay more later.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/316/1*UG0hj78c_cnPprmi2Y5TyQ.png" /></figure><h3>So, you’re working with technical debt</h3><p>It doesn’t even matter how you got to this point. It just happened and now, here you are: your schedule is a mess, releases become stressful and team morale is low.</p><p>First of all <strong>let’s see how this looks like from the perspective of your colleagues</strong>.</p><p>Technical debt makes coding a lot slower as there are areas that you need to handle carefully in order not to cause any more problems. Usually it means that the code is written poorly and, if we’re talking about a big project, it might be the case that nobody in the team even knows what exactly that piece of code does. Changes in some areas need to be done in 10 different places and whenever a bug shows up, you’re not sure what caused it and need to spend a lot of time investigating. Usually developers know where these high technical debt areas are in the project and tend to avoid them. Working on a task that requires them to make changes to that code causes <strong>frustration</strong> (“<em>oh, it’s gonna take forever”</em>) and <strong>anxiety</strong> (“<em>I have to be extra careful who knows what I’m going to break</em>”). It’s the equivalent of playing Jenga but if the tower falls then you not only have to clean everything up and put the pieces back together but you’re doing it <strong>while the pieces are on fire</strong>.</p><p>In order to make sure that the Jenga tower doesn’t fall, whenever a change is made extra testing is done by everyone in the team. Everyone, especially the tester on the team, is <strong>feeling the pressure</strong> of not letting any bugs go into production. This just makes tasks take even longer than normal and developers become “<em>duck tape programmers” </em>— trying to ship on time but desperately holding everything together behind the scenes.</p><h3>How do I manage this?</h3><p>Fist of all, I have to tell you that <strong>I lied. </strong>Remember a couple of paragraphs ago when I said that it doesn’t matter how you got here? Well, I lied, it does, because how you got here tells you how you’re going to get out.</p><p>High technical debt can have many causes :</p><ul><li>Sometimes it’s not your fault — this is an old project and you and your team “inherited” it this way or the stakeholders simply took the decision of going into debt knowing the risks because they needed fast results.</li><li>Other times, the fault is somewhere in the middle — constantly changing specs and priorities meant that you had to be flexible and adapt still, you never tried to confront the stakeholders and explain that all these fast paced changes will impact the quality of the project on the long run.</li><li>And sometimes it’s completely your fault — you’ve underestimated the work that needed to be done, the team had a lot of junior programmers and, in trying to keep up with your initial estimations, you’ve opted a little bit to often for fast results rather than good quality. Or maybe you just didn’t put that much emphasis on good engineering methodologies like code reviews.</li></ul><h4>Let’s start off with the team.</h4><p>I think the best way to manage this is to <strong>be honest</strong>. Just put it out there: “Guys, this is what we’re facing … I can see that you are frustrated, believe me so am I … Let’s see what we need to do in order to change the situation.”</p><ul><li>Have a meeting dedicated to technical debt with your team and <strong>brainstorm ideas on how to fix the situation</strong>. Try to understand how you got there and identify things that can be changed. It is also very important not to blame anyone. Start the discussion from the mindset that everyone did the best they could given the knowledge and information that they had at that specific moment.</li><li><strong>Adopt good work methodologies</strong> <strong>and stick to them</strong> (unit testing, code reviews even peer programming can work). Remember what John Wooden said: “If you don’t have time to do it right, when will you have time to do it again?”. So start working on doing things right from the start (or at least from now on).</li><li><strong>Make a wall of “Technical Debt”</strong> and put post-it notes with areas or functionalities that you know need refactoring. Start including these in your sprints one by one while still delivering business value to your stakeholders. This will also add a new sense of ownership as everyone will want to get rid of the post-it notes on that wall.</li></ul><h4>What about the stakeholders?</h4><p>Well, like any other problem it won’t get solved unless you talk to them. Be honest about the state of the project and explain to them the concept of technical debt. <strong>If stakeholders aren’t aware of the concept of technical debt, they will award fast shipping rather than quality</strong>. Once they become aware that fast comes with a cost, they might start thinking differently.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*CBX2-ADPpld-tXRJbKBABA.gif" /></figure><p>You will most probably be facing <strong>three kinds of stakeholders</strong> and the result will depend on your relationship with them:</p><ul><li><strong>Nr.1&gt;</strong> <em>The ones that knowingly took decisions that increased technical debt</em> (“We need this big functionality by next month”). You will have to prove to them that although they seem to be getting results fast, this will hit them hard on the long run and that changing specs in the middle of the sprint isn’t useful either. Practically, you have to convince them that you need to go slower or everything will fall apart. You might find that they don’t trust you (“It worked fine until now. It’s not going to crash if we do this little thing also. Don’t worry and just focus on delivering.”) so you will need data to prove your point.</li><li><strong>Nr. 2<em>&gt;</em></strong><em> The ones that took these decisions</em> (“Priorities changed, we need this by next week”) <em>without knowing the risks</em>. This happens especially when, trying to please them, you started working ASAP without addressing potential downfalls. Maybe if you told them or made sure they understand the risk, they would have realised it’s not that important or urgent and things would have been different. Depending on how bad the situation is and on how much you’ve tried (or didn’t try) to warn them, you might be facing a <strong>lack of trust</strong> (“It’s your fault, you never told us this can happen.”) or <strong>you might gain some trust</strong> (“Thanks for making us aware of this. Well, what do you suggest we do?”).</li><li><strong>Nr.3<em>&gt;</em></strong><em> The ones that had no clue that you are struggling to meet the deadlines and that quality is suffering because of that</em>. They made no changes to the initial plan but you underestimated the work that needed to be done and you ended up building the debt yourself. It probably the hardest situation to be in because you are trying not to loose the trust of the stakeholders while explaining why things are going slow.</li></ul><p><strong>Whatever the case is, you need to have a conversation and present the situation</strong>. Depending on the situation and what kind of stakeholders you have (Nr.1, Nr.2 or Nr.3) you will have to prepare differently for the conversation. Here are some things that can help:</p><ul><li>A ratio of <strong>time spent developing </strong>to<strong> time spend fixing bugs </strong>(if you think you have a problem with technical debt but you haven’t been tracking this yet, you should).<strong> </strong>Developing brings value, bug fixing doesn’t. If your team is wasting 30% of it’s time looking for and fixing bugs then you aren’t delivering value 30% of the time (translate this into money for a bigger impact). (good for Nr.1)</li><li><strong>Burndown and velocity charts</strong> showing the amount of work being done with the current state of the project. (good for Nr.1)</li><li>A short <strong>description of the state of the team</strong>. The goal here is to make them understand that if the team is frustrated or in burnout, then you won’t be able to deliver neither quality or quantity. (good for Nr.1 and Nr.2)</li><li>A list of <strong>areas that need refactoring</strong> and <strong>potential risks to the business</strong> if they are not refactored. The idea is to make the stakeholders aware that if something goes wrong in those areas, their whole business might suffer (sales going down, bad reviews etc.). (good for Nr.1 and Nr.2)</li><li><strong>Examples of bugs</strong> that almost made it live and could have had a big impact on the business. Giving a real example (“Last month we had a bug that made you unable to introduce your credit card detail if you had a discount on your items”) will make them pay more attention. (good for Nr.1 and Nr.2 and even for Nr.3 if you are showing that you are aware of the issues and are working on them)</li><li><strong>A plan</strong> regarding changes that can be made in order to improve quality (for example how you could include one area that needs refactoring in each sprint for the next 6 sprints while still delivering business value). (good for Nr.2 and Nr.3)</li><li><strong>An updated schedule</strong> showing how milestones need to change in order to develop new features and work on the technical debt at the same time. (good for Nr.2 and Nr.3)</li><li>Details on other <strong>solutions found inside the team</strong> and how this will help reduce bugs and increase quality (for example having a technical two hour meeting with the developers each sprint to discuss new functionalities and how to approach them before starting to code). (good for Nr.3)</li><li><strong>A list of what you need from them</strong>. It might be anything from putting a stop to new functionalities until you pay some of the debt, to move some of the deadlines a little bit further away or to add another developer to the team. Be sure that you are mindful of their needs as well as yours and that you find the solution that fits both. (necessary for all of them)</li></ul><h3>And now you are done with technical debt!</h3><p>I wish it would be that easy. What you should be doing now is asking yourself —<strong> “What is the next step?”. </strong>Do you have a problem and need to deal with it right now, or is this just something you should be thinking about, making sure it doesn’t happen in the future?</p><p>Before leaving I’m going to remind you just one thing: <strong>technical debt isn’t always bad, you just have to keep it under control.</strong></p><p>Oh, and here is a quote I think sums up this idea pretty well:</p><blockquote>“Shipping first time code is like going into debt. <strong>A little debt speeds development </strong>so long as it is paid back promptly with a rewrite… <strong>The danger occurs when the debt is not repaid. </strong>Every minute spent on not-quite-right code counts as <strong>interest on that debt. </strong>Entire engineering organizations can be brought to a stand-still under <strong>the debt load </strong>of an unconsolidated implementation.” — Ward Cunningham</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=679341f45065" width="1" height="1" alt=""><hr><p><a href="https://blog.devgenius.io/working-with-technical-debt-a-pm-perspective-679341f45065">Working with technical debt (a PM perspective)</a> was originally published in <a href="https://blog.devgenius.io/">Dev Genius</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>