How to vibe codeFathom

Prerequisites

• Node.jsfree

  Runtime for Next.js web application and backend scripts

• GitHubfree

  Version control and deployment pipeline source

• 

  OpenAI API Keypay-as-you-go

  Required for Whisper transcription services

AI coding tools

• Claude Code

  Best-in-class terminal agent for scaffolding full-stack apps and debugging complex asynchronous audio pipelines

• Cursor

  Ideal for fine-tuning the React dashboard UI and managing component-level edits

Build guide

1. 01Scaffold Web Application & Database Schema

   Initialize a Next.js project with Tailwind CSS, shadcn/ui, and better-auth. Set up the Neon Postgres database with pgvector extensions to store users, meetings, raw transcripts, and vector embeddings.

   Create a new Next.js 16 project with TypeScript, Tailwind CSS, and App Router. Configure better-auth for email/password authentication with a Postgres adapter. Set up a Neon Postgres connection using Drizzle ORM, defining tables for users, meetings (id, title, duration, audio_url, created_at), transcripts (id, meeting_id, speaker, text, timestamp_start, timestamp_end), and summaries (id, meeting_id, content, action_items jsonb). Ensure pgvector extension is enabled on the database schema for future embeddings storage.

   ⧉copy promptopen in Claude CodeCopen in Codexopen in Cursor

2. 02Build Audio Upload and Transcription Pipeline

   Implement a file upload interface for audio recordings (MP3/WAV/M4A) and wire up OpenAI's Whisper API to transcribe audio files into structured chunks with timestamps.

   Build a React dropzone component in the Next.js frontend allowing users to upload meeting audio files. Implement a backend API route that accepts the file, sends it to the OpenAI Whisper API with timestamp granularity enabled, and parses the response into discrete transcript objects with speaker labels, start times, end times, and text. Save these records into the transcripts table and handle file storage via Cloudflare R2 or local uploads for development.

   ⧉copy promptopen in Claude CodeCopen in Codexopen in Cursor

3. 03Implement AI Summarization and Action Item Extraction

   Create background routines or server actions that feed completed transcripts into Claude or GPT models to generate structured summaries, key moments, and extracted action items.

   Write an AI processing service using the Anthropic API (Claude Sonnet). When a transcription finishes, retrieve the full transcript text, pass it to Claude with a system prompt designed for executive meeting summaries, and instruct it to return a structured JSON object containing an executive summary, key decisions, and a list of action items with assignees. Store this output in the summaries table and display it cleanly on the meeting detail page.

   ⧉copy promptopen in Claude CodeCopen in Codexopen in Cursor

4. 04Build 'Ask Fathom' Vector RAG Search Engine

   Generate vector embeddings for meeting transcript segments, store them using pgvector, and build a conversational chat interface allowing users to query across all past meetings.

   Implement a retrieval-augmented generation (RAG) search feature ('Ask Fathom'). Write a background function that chunks meeting transcripts, generates text embeddings using OpenAI's embedding model, and stores them in a vector column in Neon Postgres. Build a chat interface API route that takes a user query, embeds it, performs a cosine similarity search across all user meetings to retrieve relevant context chunks, and streams an answer back using Claude with source meeting citations.

   ⧉copy promptopen in Claude CodeCopen in Codexopen in Cursor

5. 05Add Search, Playlists, and Export Polish

   Build global search across meetings and transcripts, highlight clipping tools, and export/share capabilities for meeting insights.

   Add a global search bar and filter interface allowing users to search meetings by keyword, attendee, or date. Create a highlight clipping feature that lets users select transcript segments to bundle into custom playlists. Implement export options to copy summaries to clipboard or send summary emails via Resend.

   ⧉copy promptopen in Claude CodeCopen in Codexopen in Cursor

Cost vs paying for Fathom