← Sonal Agarwal / Meeting Notes → Action Plan
📝

Meeting Notes → Action Plan

Paste messy meeting notes — get decisions, owned action items, and a drafted follow-up email

Paste your notes

Raw notes, a transcript, or even half-finished bullet points — Claude does the work of organizing it.

💡 This one's a live AI call, not a calculator. Claude actually reads your notes and reasons through them in multiple steps: it separates decisions from open questions, matches action items to the right owner, and drafts the email — there's no pre-written rule engine behind this one.
⚠️ Good to know Nothing is sent anywhere until you click "Generate Action Plan" and enter your own Anthropic API key. Your notes and key are used only for that one request and are never stored or saved.

Claude is reading your notes...

Separating decisions from action items, checking for gaps, drafting the follow-up...

📌 Meeting summary

🛠️ Behind this build — why and how I made this ▾

🎯 Why I built this

As a Systems & Insights Lead, I sit in a lot of meetings where decisions and action items get said out loud but never get written down properly — or they do, but the notes are a mess and nobody follows up. I wanted to build something that takes those raw, messy notes and turns them into something I could actually act on the same day: who owes what, by when, and a draft email ready to send.

What you get: a short summary, the decisions that were actually made, a list of action items with the owner and deadline pulled out (and flagged if either is missing), the open questions nobody answered, and a ready-to-send follow-up email.

🧠 How it works

Unlike my Business Acquisition Assessor (which scores things with my own rules first, and only calls Claude if you want a deeper pass), this tool is AI-only from the start — there's no rule-based fallback. The moment you click "Generate Action Plan," your notes go straight to Claude's API with a single prompt that asks it to do several distinct jobs at once: separate decisions from action items, match each action item to the right owner using the attendee list you gave it, flag anything left unassigned or without a deadline, pull out anything left unresolved, and write a follow-up email in a normal, human tone — not a generic AI-sounding one.

Claude replies in a strict structured format (JSON) that my code reads and lays out on the page. Your API key is typed into a one-time popup, used for that single request, and kept only in your browser's memory for the session — never saved or sent anywhere else.

🤖 Claude's contribution vs. mine

What Claude did
  • Wrote the prompt structure that asks for decisions, action items, open questions, and a follow-up email in one pass
  • Designed the JSON format Claude replies in and the matching parser/fallback for when it wraps the reply in extra text
  • Wrote the HTML/CSS layout, matching the visual style of my other tools
  • Suggested flagging action items with no owner or no deadline as a visible warning tag, instead of silently guessing
What I did
  • Picked this exact problem because it's something I deal with weekly, not a generic demo idea
  • Decided the tool should be AI-only with no rule-based version — this one only makes sense as a live reasoning task, not a scoring formula
  • Tested it against real, messy notes I wrote myself in meetings to check the action items actually came out right
  • Decided what "good enough" follow-up email tone looks like, and adjusted the prompt until the drafts stopped sounding stiff
  • Wrote this section, and decided how the API key should be handled, consistent with my other tools

📈 Where this sits in my Claude journey

This is my second live Claude API integration (after the Business Acquisition Assessor), and the first tool I've built that's AI-only from the first click — no rule-based version sitting underneath it. It's also a different kind of problem entirely: instead of scoring a business or a budget, it's about understanding unstructured, messy human conversation and turning it into something organized. That's a genuinely different skill than the calculators and form-based tools I built earlier.

Second live Claude API integration First AI-only tool (no rule-based fallback) First tool tied directly to my actual day job First multi-task single-prompt design

⚠️ Limitations & what's next

  • Claude can only work with what's in your notes — if something important was decided but never written down, it won't show up here.
  • It's a starting draft, not a substitute for actually checking with the people involved before you send the follow-up email.
  • Needs your own Anthropic API key, since this is a static portfolio site with no backend of its own.
  • Next on my list: letting you paste in last week's action items too, so Claude can flag anything that was promised before and still hasn't been done.