DEV Community

arafatruetbd
arafatruetbd

Posted on

How I Built an AI-Powered Transcript Summarizer Using n8n

1. Intro

Do you ever find yourself buried under mountains of meeting transcripts or interview notes? What if summarizing them could be completely automated? I just built a workflow with n8n to automatically pull transcript files from Google Drive, summarize them via an LLM, and push the results into Google Sheets—with zero manual effort.

I’m publishing the full workflow on GitHub, and in this post, I'll walk you through how it works, why I built it, and how you can apply it to your own projects.

2. Problem Statement

  • Large organizations or content creators often generate long-form transcripts—calls, podcasts, interviews, lectures.
  • Manually summarizing these is time-consuming and inconsistent.
  • Valuable insights get lost in walls of unstructured text.

3. Enter n8n + LLMs

With n8n, an open-source workflow automation tool, and modern large language models (via OpenRouter/OpenAI), you can automate:

  1. Detecting new transcript files in Google Drive
  2. Decoding and cleaning the text
  3. Chunking large transcripts into manageable pieces (≤4000 characters)
  4. Summarizing each chunk consistently into:
  • A 3-sentence executive summary
  • A list of pain points
  • A list of feature requests
    1. Merging & deduplicating chunk summaries
    2. Format results for clarity
    3. Save or update results in Google Sheets by matching CallID

No coding except for simple JavaScript cleaning—just drag, configure, and run.

4. How the Workflow Works (Step-by-Step)

  1. Google Drive Trigger
    Watches a specific Drive folder (e.g., “Transcripts”) and triggers when a new file is added.

  2. Download
    Fetches the file binary so it can be processed.

  3. Decode Binary → rawText
    Detects if the file is UTF-8 or UTF-16, strips out BOMs, and converts it into readable text.

  4. Load Transcript
    Moves the decoded text into a variable for downstream processing.

  5. Preprocessing

  • Normalizes line breaks
  • Replaces [inaudible] with “unclear audio”
  • Redacts email addresses and phone numbers for privacy
  1. Chunk Transcript
  • Splits by trimmed lines into chunks under ~4000 characters
  • Handles long lines by slicing them
  1. Basic LLM Chain Sends each chunk to the LLM with a prompt to return strict JSON with:
  • summary: exactly 3 sentences
  • pain_points[]
  • feature_requests[]
  1. Merge Summaries
  • concatenates all chunk summaries
  • deduplicates pain points & feature requests
  1. Format for Sheet

  • Generates:

    • CallID (either passed-in or generated)
    • Multi-line summary (one sentence per line)
    • Bulleted lists for pain points / feature requests
  1. Append or Update in Google Sheets
* Matches on `CallID` to avoid duplicates
* Adds or updates the row in your specified sheet
Enter fullscreen mode Exit fullscreen mode

(Add a screenshot or diagram of the workflow if available.)

5. Why This Matters

  • Save Time – No more slogging through transcripts manually
  • Consistency – A structured prompt ensures uniform output
  • Privacy – Sensitive info is redacted automatically
  • Scalable – Chunking means the solution works on long transcripts
  • Customizable – Want 5-sentence summaries or extra metadata? Just tweak the prompt or chunk code node.

6. How to Use It

  1. Clone & import the repo: 
   git clone https://github.com/arafatruetbd/n8n-transcript-summarizer.git
Enter fullscreen mode Exit fullscreen mode
  1. In n8n:
  • Workflows → Import from File, choose the JSON

  • Reconnect your credentials:

    • Google Drive
    • Google Sheets
    • OpenRouter or another LLM provider

  • Configure:

    • Drive folder ID
    • Sheet ID & GID
    • Model name if needed
  1. Activate and test. Upload a transcript to Drive and watch the summary appear in the sheet!

7. Customization Ideas

  • Add a Slack node to post summaries to a channel automatically
  • Send email summaries
  • Archive transcripts after summarizing
  • Use a different LLM (e.g., local model or Llama on a self-hosted server)
  • Include additional metadata, like speaker names, timestamps, call duration, etc.

8. Call to Action

The full workflow and README are on GitHub:
https://github.com/arafatruetbd/n8n-transcript-summarizer

Feel free to clone it, tweak it, or fork it. I’d love to see what you build on top of it or how you'd improve it—drop a comment, issue, or pull request!

Summary

This workflow turns the mundane task of summarizing transcripts into a click-and-go automation powered by n8n and LLMs. Whether you're in research, product, customer support, or content creation—if you want a smarter way to turn transcripts into insight, give this a spin.

Top comments (0)