Automated Feedback Analysis Workflow

An end-to-end no-code automation that collects customer feedback through a Google Form, classifies sentiment and generates summaries with Google Gemini AI, and writes structured results back to Google Sheets — without any manual intervention.

Built with Make.com, Google Forms, Google Gemini 2.5 Flash, and Google Sheets.

Capstone-style project from the TripleTen AI Automation program.

---

Why this project

Customer feedback is high-signal but expensive to triage by hand: someone reads each response, decides whether it's positive or negative, and writes a one-line summary so the team can act on it. This workflow replaces that loop with an event-driven pipeline that runs the moment a form is submitted, so insights land in a structured sheet ready for review or downstream reporting.

---

Architecture

Customer  ──▶  Google Form  ──▶  Make scenario  ──▶  Gemini AI  ──▶  JSON Parse  ──▶  Google Sheets
                                  (Watch Responses)   (Sentiment +    (Validate +    (Search row +
                                                       Summary)        extract)       Update row)

Stage	Module	Purpose
1	google-forms:watchResponses	Polls the form for new submissions
2	gemini-ai:createACompletionGeminiPro	Classifies sentiment and writes a one-sentence summary, returning a strict JSON object
3	json:ParseJSON	Validates and parses the model output into typed fields
4	google-sheets:filterRows	Locates the original response row in the sheet by free-text answer
5	google-sheets:updateRow	Writes Sentiment and Summary back to that row

The full flowchart (including the planned negative-sentiment Gmail alert branch) is in presentation.pdf.

---

Files in this repo

File	Purpose
README.md	This file
make-blueprint.json	Importable Make scenario (sanitized — placeholders replace personal IDs)
setup.md	Step-by-step deployment guide
gemini-sentiment-prompt.md	The prompt and output contract sent to Gemini
customer-feedback-form.md	Google Form fields and structure
google-sheets-schema.md	Column schema for the responses sheet
presentation.pdf	Project deck
LICENSE	MIT

---

Quick start

1. Clone this repo and open make-blueprint.json.
2. Create the Google Form described in customer-feedback-form.md. Note its Form ID.
3. Link a responses sheet and add the columns from google-sheets-schema.md.
4. Import the blueprint into Make.com (Scenarios → Create new → Import Blueprint).
5. Replace the placeholders (YOUR_CONNECTION_ID, YOUR_GOOGLE_FORM_ID, YOUR_GOOGLE_SHEET_ID, YOUR_GOOGLE_DRIVE_FOLDER_ID) by reconnecting your Google account and selecting your form / sheet from the dropdowns.
6. Run once manually to verify, then turn the scenario on.

Detailed walkthrough: setup.md.

---

What the AI sees and returns

The Gemini call uses a tightly-scoped prompt with a fixed output contract — see gemini-sentiment-prompt.md. The model returns:

{
  "sentiment": "Positive | Neutral | Negative",
  "summary": "One-sentence summary of the feedback."
}

Anything else fails the JSON parse step and is caught before it reaches the sheet — that's the design choice that makes this workflow safe to leave running unattended.

---

Sample results

Real submissions processed through the live scenario:

Satisfaction	Free-text feedback	Sentiment	Summary
Very satisfied	"Overall, I had a good experience. The service met my needs…"	Positive	The user was very satisfied with the experience.
Unsatisfied	"Not as expected"	Negative	The user is unsatisfied and the product did not meet expectations.
Satisfied	"Had a hands-on experience"	Positive	The user is satisfied with the hands-on experience.

---

Roadmap

Implemented in this version:

• Form-triggered ingestion
• Gemini-based sentiment + summary
• Strict JSON contract with parse-time validation
• Row-level write-back to Google Sheets

Planned next:

• Conditional Gmail alert when sentiment is Negative (branch shown in the architecture diagram)
• Slack notification fan-out
• Weekly / monthly trend reporting in a dashboard tab
• Retry + dead-letter handling for transient API errors

---

Tech

• Make.com — orchestration
• Google Forms — capture
• Google Gemini 2.5 Flash — sentiment + summary
• Google Sheets — structured output

---

License

MIT — see LICENSE.

Author

Swetha Chennuru · github.com/schennuru17-arch