grainiac-meeting-processor

name

grainiac-meeting-processor

description

Analyze a single customer meeting transcript from Grain and create or update that company's page in the Account Tracking Notion database. Use this skill when given a Grain recording ID, company name, and meeting details. The skill covers fetching the transcript, extracting information against a structured template, and writing to Notion via its API. Requires GRAINIAC_GRAIN_TOKEN, GRAINIAC_NOTION_TOKEN, and GRAINIAC_NOTION_DATABASE_ID env vars.

Grainiac Meeting Processor

Process a single meeting: fetch transcript from Grain, analyze it, create or update the company's Notion page.

Environment Requirements

• GRAINIAC_GRAIN_TOKEN — Grain personal access token
• GRAINIAC_NOTION_TOKEN — Notion integration token
• GRAINIAC_NOTION_DATABASE_ID — Account Tracking database ID (required)
• GRAINIAC_NOTION_TITLE_PROPERTY — (optional) name of the database's title property holding the company name (default: Company)

Input

The orchestrator passes these values in the prompt:

• RECORDING_ID — Grain recording UUID
• COMPANY — Company name (may need correction based on transcript content)
• GRAIN_URL — Link to the Grain recording
• MEETING_TITLE — Title of the meeting
• MEETING_DATE — Date string (YYYY-MM-DD)

Procedure

Step 1: Fetch the transcript

Use the Grain client to get the full transcript and recording metadata:

import sys
sys.path.insert(0, "scripts")
from grain_client import get_recording, get_transcript_text

recording = get_recording("<RECORDING_ID>", include_participants=True)
transcript = get_transcript_text("<RECORDING_ID>")

Also extract: title, date, participants, summary, public_url from the recording object.

Step 2: Read the full transcript

Read the transcript text in its entirety. This is the primary source of truth — not the AI summary. Extract all details that map to the Notion template categories. See references/notion-template.md for the full template.

Step 3: Analyze the transcript

Extract information for every applicable section of the template. For each category, capture verbatim quotes and specific details where possible. Key categories:

Always extract (every meeting):

• Follow-up items (action items with owner and date)
• Key activity / decision summary (one bullet for the timeline)
• People mentioned (names, titles, roles, emails, personal details)
• Sentiment and engagement signals

Extract when discussed:

• Tech stack details (specific tools, platforms, infra mentioned)
• Deployment model preferences (self-hosted, hybrid, cloud, air-gapped)
• Inference/LLM requirements (vendor-managed, BYO-LLM, gateway, air-gapped) — for AI products
• Security and compliance requirements
• Governance and admin control needs
• Integration requirements
• Feature blockers or requests
• Product use cases and target workflows
• Pricing or commercial discussions
• Procurement/legal status (NDA, legal review, vendor onboarding)
• Communication preferences (Slack, email, cadence)

Important analysis guidelines:

• Distinguish between what the customer said vs. what our team said.
• Capture customer concerns and objections verbatim where possible.
• Note buying signals (proactive next steps, urgency, bringing in stakeholders).
• Flag corrections to previously recorded information.
• Mark sections as "Not discussed" when a topic did not come up — do not fabricate.

Step 4: Check for existing Notion page

from notion_client import find_company_page
page = find_company_page("<COMPANY>")

If page is None, proceed to Step 5a (create). Otherwise, proceed to Step 5b (update).

Step 5a: Create a new page

Build the full page structure following the template in references/notion-template.md. Use the block builders from scripts/notion_client.py:

from notion_client import (
    create_page, append_blocks, rich, heading1, heading2, bullet, todo, divider, paragraph
)

The page sections, in order:

1. Follow-Up Items — checkbox list of action items from our team
2. Key Activities & Decisions — reverse-chronological milestones
3. Meeting Log — hyperlinked meeting titles, newest first
4. Company Overview — name, industry, size, deal stage, sentiment
5. People & Organization — stakeholders, org structure
6. Tech Stack & Environment — dev env, source control, infra, auth
7. Current Tooling & Competitive Landscape
8. Requirements — deployment model, inference/LLM, security, governance, integrations, blockers
9. Use Cases — primary and secondary use cases, priority ranking
10. Commercial — pricing, procurement/legal, contract
11. POC / Pilot Tracker
12. Communication — channels, cadence, scheduling
13. Sentiment & Relationship Notes

Separate each H1 section with a divider block. Notion limits 100 blocks per API call — use multiple append_blocks calls for the remaining blocks after the initial create_page.

Step 5b: Update an existing page

This is the more common case. Fetch existing blocks and surgically update:

from notion_client import (
    get_all_blocks, append_blocks, update_block,
    find_heading_block, get_last_block_in_section,
    rich, bullet, todo
)

page_id = page["id"]
blocks = get_all_blocks(page_id)

Update strategy for each section:

Follow-Up Items: Add new todo blocks after the last existing todo in the section. If a previous follow-up was completed (confirmed in transcript), update its block with checked: True.

Key Activities & Decisions: Insert a new bullet after the last existing activity bullet (before the divider). Use format: **YYYY-MM-DD** — Summary of key decision or milestone.

Meeting Log: Insert the new meeting as the first bullet after the "Meeting Log" heading (newest first). Format: hyperlinked title.

All other sections: For each piece of new information:

• If it supplements existing content, append new bullets after the last block in that section.
• If it corrects existing content, use update_block to modify the specific block.
• If a section previously said "Not discussed" and now has data, replace that block.

Use find_heading_block and get_last_block_in_section to locate insertion points. Use find_block_by_text to find specific blocks to update.

Step 6: Verify

After all writes, fetch the blocks again and do a sanity check:

• Confirm the new meeting appears in the Meeting Log
• Confirm new follow-up items were added
• Confirm the key activity was logged

Report completion with: company name, meeting title, whether page was created or updated, and the Notion page URL.

Important Notes

• Notion API limits: 100 blocks per append call, 3 requests/second. Batch blocks and add brief delays if needed.
• All block text content has a 2000-character limit per rich text segment. Split long content across multiple segments.
• When the transcript is ambiguous or the company name from email domain seems wrong (e.g., "Gmail"), use context from the transcript to determine the correct company name.
• For the Meeting Log entry, the Grain URL should come from the recording's public_url field.