Productivity~8 hours to build$1K/Month goal

AI Meeting Notes Cleaner

Build an AI tool that transforms messy Zoom and Teams transcripts into clean summaries, decisions, and action items with owners and due dates.

The Problem

Remote teams drown in meeting artifacts. Zoom and Microsoft Teams ship auto-transcripts full of false starts, crosstalk, and “can you hear me?”—but turning that blob into accountable action items still takes 20–30 minutes per call. Otter and Fireflies record everything yet overwhelm users with word dumps; managers want a clean decision log, not another searchable archive nobody opens.

Product leads describe the same failure mode: “We have notes in three places and still miss follow-ups.” Granola and Fathom nail live capture for executives, but indie teams and consultants need a weekend-buildable cleaner—upload transcript, pick output format (standup summary, client recap, RACI actions), export to Notion or email. No bot joining calls required for v1.

The Solution

MVP: paste-or-upload transcript UI, LLM pipeline that extracts attendees, decisions, risks, and action items with suggested owners, plus one-click Markdown/Notion export. Optional Zoom webhook in v2. Focus on clarity over storage—one screen in, structured doc out.

How it works:

Ingest — Paste transcript or upload .vtt from Zoom/Teams; detect speakers and timestamps
Clean — LLM removes filler, merges duplicate threads, tags decisions vs brainstorm
Structure — Output sections: Summary, Decisions, Action items (owner, due), Open questions
Export — Copy Markdown, send email recap, or push to Notion via API

Market Research

Meeting intelligence is crowded at the top but thin in the “transcript janitor” wedge for small teams:

AI meeting assistants — MarketsandMarkets projects the broader conversational AI segment exceeding $40B by 2027; meeting notes are a high-frequency wedge.
Remote work persistence — Gallup data shows over 20% of US workers still hybrid/remote, sustaining demand for async recap quality.
Transcript volume — Zoom reported 300M+ daily meeting participants at peak; even a tiny share exporting transcripts creates massive TAM for post-processing tools.
Wedge — Compete on speed and format templates (client-facing vs internal), not on joining calls as a bot.

Competitive Landscape

Incumbents capture recordings; you win on post-processing simplicity and template depth:

Otter.ai — Live transcription + search. Strong for lectures; action extraction and client-ready summaries need manual editing. Free tier · Pro ~$17/user/mo · Business ~$30/user/mo
Fireflies.ai — Bot joins calls, CRM integrations. Heavy for consultants who just want a cleaned recap from an existing .vtt file. Free · Pro $18/user/mo · Business $29/user/mo
Granola — Mac-native live notes with AI enhancement. Beautiful UX; limited if you already have Zoom transcript only. Free beta · paid tiers emerging
Fathom — Free AI notetaker for Zoom with highlight clips. Great for sales calls; less flexible custom output schemas for ops teams. Free core · Team plans for CRM sync

Your Opportunity

Be the “PDF compressor” of meeting notes: one job, instant value, $12/mo for unlimited cleans. Win SEO on “clean Zoom transcript” and “meeting notes to action items.”

Business Model

Usage-based free tier drives viral loops inside Slack communities. Target $1k MRR at ~85 Pro users or a mix of Pro + small team plans.

Free ($0) — 5 cleans/month, watermark on export
Pro ($12/mo) — Unlimited cleans, custom templates, Notion export
Team ($39/mo) — 5 seats, shared template library, SSO later

Unit economics

~$0.08–0.15 LLM cost per clean on GPT-4o-mini class models · Pro margin >80% · CAC via template gallery and “paste your transcript” demo on landing page.

Recommended Tech Stack

Keep v1 serverless: upload → queue → LLM → structured JSON → render. No call bot infrastructure required.

Next.js 14 App Router — Upload UI, streaming results, auth via Clerk or Supabase.
OpenAI GPT-4o-mini / Claude Haiku — Structured output with JSON schema for action items; retry on validation failure.
Supabase — Store transcripts, outputs, template presets; RLS per user.
Vercel + Edge — Deploy API routes; optional Inngest for async long transcripts.
Notion API — OAuth export as killer Pro feature; map sections to page blocks.
Stripe Billing — Metered or flat Pro subscription; usage caps on free tier.

AI Prompts to Build This

Copy and paste these into Claude, Cursor, or your favorite AI tool.

1. Project Setup

Scaffold Next.js 14 TypeScript app “NotesCleaner” with Supabase auth, table meeting_cleans (raw_text, cleaned_json, template_id), and Zod schema for output: { summary, decisions[], actions[{owner, task, due}], open_questions[] }. Add /api/clean POST route calling OpenAI with structured outputs.

2. Core Feature

Build transcript paste UI with character count, template picker (Standup, Client recap, Board notes), streaming markdown preview, and speaker attribution heuristic from “Name:” patterns. Include edit-in-place before export and copy-to-clipboard.

3. Landing Page

Landing page: interactive demo with sample messy transcript pre-loaded, before/after toggle, comparison table vs Otter/Fireflies (we don't join your calls), pricing, FAQ on privacy (transcripts deleted after 30 days). Hero: “Your meeting ended. Your follow-ups didn't.”

4. Notion Export

Implement Notion OAuth and block mapper: Summary → callout, Decisions → bulleted list, Actions → to-do blocks with @mentions parsed from attendee list.