<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: Touhid</title>
    <description>The latest articles on DEV Community by Touhid (@touhid314).</description>
    <link>https://dev.to/touhid314</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4255807%2Fa0eff9d8-8a97-4169-a4d4-2583b46b6fe5.png</url>
      <title>DEV Community: Touhid</title>
      <link>https://dev.to/touhid314</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/touhid314"/>
    <language>en</language>
    <item>
      <title>Building a live AI tutor that teaches on a blackboard</title>
      <dc:creator>Touhid</dc:creator>
      <pubDate>Fri, 20 Nov 2026 13:39:54 +0000</pubDate>
      <link>https://dev.to/touhid314/building-a-live-ai-tutor-that-teaches-on-a-blackboard-5cfj</link>
      <guid>https://dev.to/touhid314/building-a-live-ai-tutor-that-teaches-on-a-blackboard-5cfj</guid>
      <description>&lt;p&gt;A recorded lesson has a fixed next step. When an explanation stops making sense, the learner has to pause, search elsewhere, or continue without understanding it. I wanted to explore a different interaction: a tutor that can explain visually and respond when the learner interrupts.&lt;/p&gt;

&lt;p&gt;That is the idea behind MyLecture, a desktop app for Windows and Mac. Its live AI tutor speaks while writing, drawing and animating on a blackboard. Learners can ask follow-up questions, bring an image into the lesson, and answer practice checks. The app also includes a studio for narrated animated videos, but the tutoring interaction starts with a smaller goal: make the confusing step discussable.&lt;/p&gt;

&lt;h2&gt;
  
  
  A research motivation, with a boundary
&lt;/h2&gt;

&lt;p&gt;Benjamin Bloom's 1984 paper, &lt;a href="https://journals.sagepub.com/doi/10.3102/0013189X013006004" rel="noopener noreferrer"&gt;The 2 Sigma Problem&lt;/a&gt;, discussed studies comparing conventional instruction, mastery learning and tutoring. The average student in the tutoring condition scored about two standard deviations above the conventional-class average, corresponding to performance above roughly 98% of that comparison group. The tutoring condition included formative checks and corrective feedback. The &lt;a href="https://gwern.net/doc/psychology/1984-bloom.pdf" rel="noopener noreferrer"&gt;original paper's first page&lt;/a&gt; describes the setup.&lt;/p&gt;

&lt;p&gt;That finding concerns particular human-tutoring studies. It does not show that an AI tutor, a blackboard interface, or MyLecture produces the same result. I have not established that outcome for this app. For me, the useful design question is how software can make explanation, questions, practice and correction available in a more responsive interaction.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why the blackboard matters to the interaction
&lt;/h2&gt;

&lt;p&gt;A block of text can contain a correct explanation while still making relationships hard to see. A diagram gives the tutor something to point at. Animation can show a relationship changing over time. Writing and speech can develop an explanation together instead of asking the learner to translate between a paragraph and a separate picture.&lt;/p&gt;

&lt;p&gt;The current tutor walkthrough uses an eclipse lesson. The explanation includes an image and a moving Sun–Moon–Earth diagram. This is a concrete example of the interaction I am building: the learner can watch the relationship being explained, then ask about a particular part of it.&lt;/p&gt;

&lt;p&gt;Visuals are not automatically good teaching. An animation can distract, a diagram can be misleading, and spoken narration can get ahead of what is visible. The useful question is whether a learner can connect the explanation to the thing on the board. That needs observation and feedback, rather than an assumption that more movement means more understanding.&lt;/p&gt;

&lt;h2&gt;
  
  
  Starting as a learner
&lt;/h2&gt;

&lt;p&gt;The app is free, but the connected AI service is separate. A learner connects their own ChatGPT, Claude or Gemini plan, or an API key. That provider's limits and charges apply. Running the desktop interface on a personal computer does not make the connected model offline or remove the need for a network connection.&lt;/p&gt;

&lt;p&gt;After connecting a provider, a useful first request is a small, specific topic with some indication of what is confusing. For example: explain why an eclipse does not happen every month, using a diagram. This is a suggested learning prompt, rather than a promise about an exact scripted response.&lt;/p&gt;

&lt;p&gt;The tutor then has a visual surface for its explanation. As a learner, I would begin with one question and watch whether the explanation addresses it, instead of requesting an entire course before checking the first concept. A short initial exchange makes it easier to notice when the explanation has taken a wrong direction.&lt;/p&gt;

&lt;h2&gt;
  
  
  Interrupting is part of the lesson
&lt;/h2&gt;

&lt;p&gt;MyLecture supports asking a question during a lesson and receiving a fresh visual explanation. That changes how a learner can use the board: an unclear diagram can become the subject of the next exchange, rather than something to silently work around.&lt;/p&gt;

&lt;p&gt;For instance, a learner might ask for a slower explanation of one relationship, or ask how a newly introduced term connects to the picture. These are examples of ways to use the conversation, not claims that the app detects confusion automatically. The learner still needs to signal what they want clarified.&lt;/p&gt;

&lt;p&gt;Images provide another starting point. A learner can attach a picture and ask the tutor to explain it. A useful question identifies the region or concept of interest. A poor-quality image or an ambiguous prompt can lead to an incorrect interpretation, so the learner should check that the tutor is explaining the intended object before building on the answer.&lt;/p&gt;

&lt;h2&gt;
  
  
  Practice creates another opportunity to correct
&lt;/h2&gt;

&lt;p&gt;The tutor can present multiple-choice questions and give feedback on an answer. That adds a moment where the learner has to choose, rather than only watch. Follow-up explanation can then address the response.&lt;/p&gt;

&lt;p&gt;An immediate correctness label is not proof of understanding. A learner can guess correctly, and AI-generated questions or feedback can contain errors. I would use a practice check as a prompt to explain the reasoning: why does the chosen answer follow from the diagram? For important coursework, checking against trusted course material remains useful.&lt;/p&gt;

&lt;p&gt;The next product question is whether these interactions help people learn and return. A completed lesson or an exported file alone does not answer that. Evidence about understanding needs an appropriate evaluation, and I do not want an attractive walkthrough to stand in for it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where the video studio fits
&lt;/h2&gt;

&lt;p&gt;MyLecture also lets creators turn an idea or notes into a narrated animated MP4. The studio supports longer videos, short reels, product demos and explainers. Creators can change a voice, edit a narration line and regenerate it, choose a theme or customize one, and set the export format.&lt;/p&gt;

&lt;p&gt;That is a different workflow from live tutoring. A video has a planned sequence that viewers can replay; a live lesson can respond to a question. Landscape, portrait and square exports, together with resolution, frame-rate and caption controls, help the creator choose how to present the finished video.&lt;/p&gt;

&lt;p&gt;MyLecture is free for Windows and Mac: &lt;a href="https://github.com/touhid314/mylecture/releases/latest" rel="noopener noreferrer"&gt;download the latest release&lt;/a&gt; (more at &lt;a href="https://mylecture.ai" rel="noopener noreferrer"&gt;mylecture.ai&lt;/a&gt;). Our walkthrough videos show the real app with editorial changes, including removed waiting time; they are demonstrations of interaction, not generation-speed measurements. If you try the tutor, the most useful feedback is specific: which explanation helped, which step was confusing, and what question you asked next.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Disclosure: I'm the maker of this project. This post was drafted with AI assistance and edited for accuracy against the project's code and documentation.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>edtech</category>
      <category>learning</category>
      <category>showdev</category>
    </item>
    <item>
      <title>Leader-based vs orchestrator-based multi-agent coding</title>
      <dc:creator>Touhid</dc:creator>
      <pubDate>Fri, 20 Nov 2026 13:37:59 +0000</pubDate>
      <link>https://dev.to/touhid314/leader-based-vs-orchestrator-based-multi-agent-coding-2lko</link>
      <guid>https://dev.to/touhid314/leader-based-vs-orchestrator-based-multi-agent-coding-2lko</guid>
      <description>&lt;p&gt;I built Super Agent because I wanted to work with several coding agents inside the project I was already editing. The question that interested me was practical: who should own a task when more than one agent is available, and how much coordination should happen before anyone changes a file?&lt;/p&gt;

&lt;p&gt;Super Agent is a VS Code extension that brings coding agents such as Claude Code, Codex and Antigravity into one workspace. One agent leads the request, works on it, and delegates independent pieces to available teammates. I describe that as leader-based multi-agent coding. The phrase describes a workflow choice, rather than a claim that coordination has disappeared.&lt;/p&gt;

&lt;h2&gt;
  
  
  What I mean by a working lead
&lt;/h2&gt;

&lt;p&gt;An orchestrator-based design can give a dedicated manager responsibility for choosing workers, distributing context, and collecting results. That can be a useful separation of responsibilities, especially when a workflow needs explicit stages or tightly controlled routing.&lt;/p&gt;

&lt;p&gt;In Super Agent, the agent responsible for the answer is also a coding agent doing the work. There is no separate dedicated manager model above it. The lead can inspect the project, make changes, and give a teammate a bounded task when that task is worth separating.&lt;/p&gt;

&lt;p&gt;These are not universal opposites. A manager can have implementation tools, and a working lead still orchestrates when it delegates. The meaningful question is where responsibility sits in this particular product, and whether the resulting workflow fits the job you have.&lt;/p&gt;

&lt;h2&gt;
  
  
  Choosing something worth delegating
&lt;/h2&gt;

&lt;p&gt;Imagine a request to change an installation option and update the setup guide. The lead can investigate and implement the option while a teammate checks the documentation and prepares the corresponding edit. Both need an agreed description of the new behavior, and the lead still needs to review the combined result.&lt;/p&gt;

&lt;p&gt;Now consider a small bug whose cause is unclear. Sending one agent to fix it and another to write tests immediately may duplicate investigation or lock in incompatible assumptions. It can be more useful for one agent to understand the failure first, then delegate a specific follow-up once the behavior is clear.&lt;/p&gt;

&lt;p&gt;The important boundary is independence. Separate filenames are helpful, but not sufficient: two files can embody the same unresolved design decision. A useful handoff explains the desired result, relevant constraints, which files the teammate owns, and what evidence should come back. Those details are part of the work.&lt;/p&gt;

&lt;h2&gt;
  
  
  Sharing a workspace changes the problem
&lt;/h2&gt;

&lt;p&gt;Super Agent's teammates work in the same project folder. That makes their changes available together, but it also means coordination matters. Two agents editing the same function can interfere with each other. A teammate can validate against a file that changes before the lead reviews it.&lt;/p&gt;

&lt;p&gt;My preferred task shape gives each teammate a clear area and keeps the dependent work with the lead until the dependency is resolved. When results return, the lead needs to inspect the actual changes and run appropriate checks on the combined state. A confident completion message from a teammate is useful context; it is not a substitute for verification.&lt;/p&gt;

&lt;p&gt;That is also why I avoid treating a larger team as an automatic upgrade. A narrow change may be easiest to finish in a single session. More independent investigations can help a broad task, but more agents also produce more decisions to reconcile.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Telegram workflow
&lt;/h2&gt;

&lt;p&gt;The other part of Super Agent is access to that workstation through Telegram. After pairing a bot, I can send requests, see progress, respond to permission requests, and receive outputs such as screenshots and videos. Text, photos, files, and voice notes are supported inputs.&lt;/p&gt;

&lt;p&gt;A practical example is asking for a preview while away from the keyboard, then reading the result on the phone and sending a correction. Permission prompts remain decisions that need a response. Being able to reach the workspace remotely does not make an ambiguous request precise or make an unreviewed change ready to release.&lt;/p&gt;

&lt;p&gt;“Always on” has a specific deployment meaning here. Super Agent can run inside VS Code on a Windows or Linux workstation. On Linux, it can also run as a standalone service with VS Code closed. The computer must stay running and connected. Super Agent does not include cloud hosting, and shutting down the machine stops that workflow.&lt;/p&gt;

&lt;p&gt;Voice notes are transcribed locally with Whisper before their text reaches the coding workflow. That is a statement about speech transcription, not a claim that the connected coding models run locally or that all communication stays offline.&lt;/p&gt;

&lt;h2&gt;
  
  
  Context transfer has limits
&lt;/h2&gt;

&lt;p&gt;When work moves between supported agents, transcript handover helps the next agent understand what happened. The project documents different transfer paths for same-CLI sessions and cross-CLI handovers. Inline material is bounded, and a transcript file can carry additional history. See the &lt;a href="https://github.com/touhid314/super-agent" rel="noopener noreferrer"&gt;project documentation&lt;/a&gt; for the current behavior.&lt;/p&gt;

&lt;p&gt;History does not ensure identical interpretation. A successor can miss a decision, overlook a file, or reason differently about the same evidence. Keeping relevant decisions explicit and checking the resulting work remains necessary. I would rather describe the information being transferred than promise that context can never be lost.&lt;/p&gt;

&lt;h2&gt;
  
  
  What this design does and does not establish
&lt;/h2&gt;

&lt;p&gt;Removing a separate manager model does not remove delegation costs. A teammate consumes calls and context. The lead spends time defining work, reading reports, resolving differences, and checking the result. For some jobs that work is worthwhile; for others it is extra effort without enough benefit.&lt;/p&gt;

&lt;p&gt;I do not have a controlled benchmark establishing a general speed or cost advantage over another architecture. A useful evaluation would compare the same task and acceptance criteria, including failures and review effort. Until then, the workflow is something to assess on your own work rather than a performance guarantee.&lt;/p&gt;

&lt;p&gt;Super Agent is a free extension under CC BY-NC 4.0. You bring your own agent accounts, and provider rules and limits apply. Start with one agent, try a task with an independent second piece, and examine whether the handoff helped. You can install it from the &lt;a href="https://marketplace.visualstudio.com/items?itemName=touhidul-islam.super-agent-touhid" rel="noopener noreferrer"&gt;VS Code Marketplace&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;Source and docs: &lt;a href="https://github.com/touhid314/super-agent" rel="noopener noreferrer"&gt;github.com/touhid314/super-agent&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Disclosure: I'm the maker of this project. This post was drafted with AI assistance and edited for accuracy against the project's code and documentation.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>ai</category>
      <category>vscode</category>
      <category>productivity</category>
      <category>showdev</category>
    </item>
  </channel>
</rss>
