The latest articles on DEV Community by Dishant Gotis (@dishantgotis). https://dev.to/dishantgotis https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3624055%2F8c4f9c65-0f2a-407f-8740-bca372bec589.jpg https://dev.to/dishantgotis en Dishant Gotis Sat, 18 Apr 2026 20:38:47 +0000 https://dev.to/dishantgotis/just-cc-this-ai-it-will-schedule-your-meeting-for-you-4882 https://dev.to/dishantgotis/just-cc-this-ai-it-will-schedule-your-meeting-for-you-4882 <h1> MailMind: An AI Agent That Actually Handles Email Scheduling </h1> <h2> Why I built this </h2> <p>Scheduling meetings over email is surprisingly painful.</p> <p>You send one mail, then it turns into a long thread:<br><br> “Are you free Tuesday?”<br><br> “No, what about Wednesday?”<br><br> “Let’s do next week.”<br><br> …and it just keeps going.</p> <p>The problem isn’t email itself — it’s that we’re using an unstructured system to solve a structured task.</p> <p>Most tools try to fix this, but they all require behavior change:</p> <ul> <li>Calendly → click a link </li> <li>AI assistants → draft replies, but don’t act </li> <li>Bots → break in real conversations </li> </ul> <p>I wanted to try something different:<br><br> <strong>what if the system just worked inside email, without changing how people use it?</strong></p> <h2> What MailMind does </h2> <p>MailMind is an AI agent that joins an email thread and handles scheduling end-to-end.</p> <p>You just CC it once.</p> <p>From there, it:</p> <ul> <li>reads the conversation </li> <li>understands availability from natural language </li> <li>follows up if something is unclear </li> <li>finds a common time </li> <li>creates a calendar event </li> <li>sends confirmations </li> </ul> <p>No links. No UI. No extra steps.</p> <p>Everything happens inside the same email thread.</p> <h2> How it actually works </h2> <p>At a high level, the system is event-driven.</p> <p>An incoming email triggers the entire pipeline:</p> <ul> <li>Gmail pushes an event </li> <li>backend fetches full thread data </li> <li>agent processes it </li> <li>system decides what to do next </li> </ul> <p>Instead of writing this as a linear script, I used a <strong>state-based agent loop</strong>.</p> <p>Each step in the flow is a node:</p> <ul> <li>classify the email </li> <li>extract availability </li> <li>check if something is missing </li> <li>compute overlaps </li> <li>generate a response </li> </ul> <p>The agent moves between these states depending on what it sees in the thread.</p> <h2> LLM + deterministic logic </h2> <p>The LLM is not running the whole system.</p> <p>It’s only used where natural language is involved:</p> <ul> <li>understanding emails </li> <li>detecting ambiguity </li> <li>rewriting responses </li> </ul> <p>Everything else is deterministic:</p> <ul> <li>time parsing </li> <li>overlap calculation </li> <li>slot ranking </li> <li>calendar operations </li> </ul> <p>This separation makes the system more reliable.</p> <h2> Maintaining context </h2> <p>Each email thread is treated as a persistent session:</p> <ul> <li>who has replied </li> <li>what time slots they gave </li> <li>who is pending </li> <li>preferences (like “no Fridays”) </li> </ul> <p>This state is stored in SQLite and updated after every email.</p> <h2> Choosing the best time </h2> <p>A scoring function ranks slots based on:</p> <ul> <li>attendance </li> <li>preferences </li> <li>priority participants </li> <li>time optimization </li> </ul> <p>The system selects the most practical option, not just the first valid one.</p> <h2> Safety layer </h2> <p>Before sending any email:</p> <ul> <li>draft goes to Telegram </li> <li>approve / reject / edit </li> <li>auto-send fallback </li> </ul> <h2> Tech Stack </h2> <ul> <li>Python </li> <li>FastAPI </li> <li>LangGraph </li> <li>Gemini 2.0 Flash </li> <li>SQLite </li> <li>React </li> <li>Gmail API </li> <li>Google Calendar API </li> <li>Telegram Bot API </li> </ul> <h2> Final thought </h2> <p>Most AI tools assist.</p> <p>This one acts.</p> <p>It takes something messy (email threads) and turns it into a structured, executable workflow.<br> <a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F0pnlihvkpkvau3pzk57r.jpeg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2F0pnlihvkpkvau3pzk57r.jpeg" alt=" " width="716" height="1600"></a><br> <a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fap9gtp9m5j5kz27cv9rp.jpeg" class="article-body-image-wrapper"><img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Farticles%2Fap9gtp9m5j5kz27cv9rp.jpeg" alt=" " width="716" height="1600"></a><br> That’s where AI becomes actually useful.</p> agents ai productivity showdev