{"slug": "i-built-a-meeting-copilot-that-remembers-what-everyone-said", "title": "I Built a Meeting Copilot That Remembers What Everyone Said", "summary": "A developer built Fathom Rebuild, an open-source AI meeting copilot that turns recorded meetings into transcripts, summaries, decisions, action items and highlights, and lets users ask questions across all their meetings with answers cited back to the source transcript. The app uses a LangGraph and LangChain agentic workflow with Groq for LLM inference, built on Next.js 15 and TypeScript, with the live meeting-capture layer (Zoom/Meet/Teams bot) intentionally stubbed out. The project is fully populated with seed data on first load and requires a Groq API key only for AI generation.", "body_md": "*This is a submission for the [Hacktoberfest Weekend Challenge: Build for a Friend](https://dev.to/challenges/hacktoberfest-weekend-2026-10-01)*\n\nI built **Fathom Rebuild**, an AI meeting copilot for a friend who spends a lot of time in meetings and often needs to go back through transcripts to find specific decisions, action items, or things someone mentioned.\n\nInstead of manually searching through long meeting transcripts, the app lets you ask questions across your meetings and get answers with references back to the relevant transcript.\n\nIt has two main workflows:\n\nThe goal was simple: **make meeting history something you can actually talk to instead of something you have to search through.**\n\n🎥 **Walkthrough:**\n\n🚀 **Deployed App:**\n\nThe project is completely open source:\n\n🔗 **GitHub:**\n\nA Fathom-style meeting assistant built for a 24-hour assignment: record → transcript → AI summary, decisions, action items and highlights, plus search and \"ask across all meetings\" with cited answers.\n\n**Stack:** Next.js 15 (App Router) · TypeScript · Tailwind CSS + shadcn/ui · LangChain + **LangGraph** · **Groq** · optional Supabase/Postgres.\n\nThe real meeting-capture layer (Zoom/Meet/Teams bot) is **intentionally stubbed**. See [Capture layer](https://github.com/Adii0906/fathom-rebuild#capture-layer-intentionally-stubbed).\n\nPrerequisites: **Node 20.9+** (Node 22 recommended) and a free [Groq API key](https://console.groq.com/keys).\n\n```\ngit clone https://github.com/Adii0906/fathom-rebuild && cd fathom-rebuild\ngit checkout claude/happy-hopper-jytoyu      # until this is merged to main\n\nnpm install\n\ncp .env.example .env.local                   # then edit it\n#   GROQ_API_KEY=gsk_...                     (required for AI generation, never commit this)\n\nnpm run dev                                  # http://localhost:3000\n```\n\nThe app is fully populated on first load, because the seed data needs no API key or database. `GROQ_API_KEY` is only used when you:\n\nThe repository contains the application code as well as the agent development logs.\n\nI built the application around an agentic workflow using **LangGraph** and **LangChain**, with **Groq** for LLM inference.\n\nThe architecture is split into separate workflows for different tasks instead of putting everything into one large prompt.\n\nThe main flow looks roughly like:\n\n**Transcript → Processing → Analysis → Structured Information**\n\nand for cross-meeting questions:\n\n**Question → Retrieval → Relevant Meeting Context → Agent Reasoning → Cited Answer**\n\nI also added references to the source transcript so the user can trace an answer back to where the information came from.\n\nThe goal was to use the AI agent for actual reasoning and workflow orchestration rather than simply sending a transcript to an LLM and asking it to summarize.\n\nOpen innovation made it possible for me to build the entire agent workflow around tools and frameworks that I could inspect, modify, and experiment with.\n\nUsing open-source frameworks like **LangGraph** and **LangChain** meant I could control how the different AI steps were connected instead of treating the intelligence as a black box.\n\nIt also makes the project easier for someone else to extend — whether that's adding new meeting analysis nodes, changing the retrieval strategy, adding another model provider, or building completely new workflows on top of the existing system.", "url": "https://wpnews.pro/news/i-built-a-meeting-copilot-that-remembers-what-everyone-said", "canonical_source": "https://dev.to/spliot_s_0d00efb657433a33/i-built-a-meeting-copilot-that-remembers-what-everyone-said-3189", "published_at": "2026-10-02 05:03:22+00:00", "updated_at": "2026-10-02 05:14:26.593055+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "large-language-models", "developer-tools", "ai-products"], "entities": ["Fathom Rebuild", "LangGraph", "LangChain", "Groq", "Next.js", "Supabase", "GitHub", "Hacktoberfest"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/i-built-a-meeting-copilot-that-remembers-what-everyone-said", "markdown": "https://wpnews.pro/news/i-built-a-meeting-copilot-that-remembers-what-everyone-said.md", "text": "https://wpnews.pro/news/i-built-a-meeting-copilot-that-remembers-what-everyone-said.txt", "jsonld": "https://wpnews.pro/news/i-built-a-meeting-copilot-that-remembers-what-everyone-said.jsonld"}}