Sameer Shah

Posted on Apr 13

Building an AI-Powered Event Feedback System: Gemini 2.5 + FastAPI + Supabase + Automated PDF Reports

#ai #fastapi #nextjs #automation

After running an event, the feedback collection process is almost always broken. You get a spreadsheet of raw responses, manually tally the ratings, write a summary email to the organizer, and hope you didn't miss anything important. It's tedious, slow, and doesn't scale.

I built a full-stack AI-powered event feedback system that automates this entire pipeline end-to-end — from the moment an attendee submits feedback to a branded PDF report landing in the organizer's inbox, complete with AI-generated sentiment analysis and improvement suggestions.

What the System Does

Every feedback submission triggers a fully automated chain:

Form data is cleaned and validated by the FastAPI backend
Gemini 2.5 analyzes the text for sentiment, urgency, highlights, and suggestions
A branded HTML report is generated and converted to PDF
The PDF is emailed to the event organizer via Gmail SMTP
All data (including AI output) is persisted to a Supabase PostgreSQL database
The analytics dashboard reflects the new submission in real time

System Architecture

Form Submission
      │
      ▼
Data Cleaning & Normalization
      │
      ▼
Gemini 2.5 AI Analysis
      │
      ▼
HTML Report Generation (Branded)
      │
      ▼
PDF Conversion
      │
      ├──▶ SMTP Email to Organizer
      │
      └──▶ Supabase Database Storage
                │
                ▼
      Dashboard & Analytics
                │
                ▼
      Date-Range Bulk Report (PDF + Email)

Tech Stack

Layer	Technology
Frontend	Next.js + Tailwind CSS
Backend	FastAPI (Python)
AI Analysis	Google Gemini 2.5
Email	Gmail SMTP (TLS, Port 587)
PDF Generation	HTML-to-PDF conversion
Database	Supabase (PostgreSQL)

The AI Analysis Layer: Gemini 2.5

The most interesting part of this system is the AI pipeline. Rather than using a generic sentiment library, I integrated Google Gemini 2.5 directly to generate structured, actionable intelligence from each feedback submission.

The prompt is designed to return a consistent JSON structure every time:

{
  "sentiment": "Positive | Neutral | Negative",
  "summary": "2-3 sentence summary of the feedback",
  "highlights": ["key point 1", "key point 2"],
  "improvementSuggestions": ["suggestion 1", "suggestion 2"],
  "urgency": "Low | Medium | High"
}

The urgency field is particularly useful — it lets organizers immediately identify submissions that need follow-up action, without reading every comment manually.

Tip: When prompting Gemini for structured JSON output, always explicitly specify the expected schema in your prompt and validate the response server-side. LLMs can occasionally deviate from the schema under edge cases.

📄 Automated PDF Report Generation

After AI analysis, the backend generates a branded HTML report and converts it to PDF. The report includes:

Event metadata (name, date, organizer)
Attendee information
Star rating visualization
AI-generated summary and highlights
Improvement suggestions with urgency indicator
Submission ID and timestamp for traceability

Using HTML-to-PDF conversion rather than a PDF library gives you full control over visual design — CSS, fonts, colors, layout — without being constrained by a library's API.

SMTP Email Delivery

The PDF report is automatically emailed to the event organizer via Gmail SMTP:

# Email config
SMTP_HOST = "smtp.gmail.com"
SMTP_PORT = 587
SMTP_SECURITY = "TLS"

# The SMTP password is stored in .env — never hardcoded

Important: Gmail requires an App Password (not your account password) when authenticating via SMTP with 2FA enabled. Store it exclusively in environment secrets.

Database Schema: Supabase

All submissions and AI outputs are persisted to a Supabase PostgreSQL table (feedback_reports). The schema captures everything — raw input, AI analysis fields, PDF URL, and email delivery status — in a single row per submission, making dashboard queries simple and fast.

Key fields: submissionId, eventName, eventDate, rating, sentiment, summary, urgency, highlights, improvementSuggestions, pdfUrl, emailSent.

The Analytics Dashboard

The Next.js dashboard connects to the API and provides:

Real-time submission counts and average ratings
Sentiment breakdown (Positive / Neutral / Negative)
High-urgency submission highlighting
Filterable table with date range, event name, and sentiment filters
Per-submission PDF access links

The bulk report feature is especially powerful: select a date range, and the system retrieves all submissions in that window, passes the full dataset to Gemini for a consolidated analysis, generates a combined PDF, and emails it to the admin. A multi-hour manual reporting task reduced to a single button click.

Security Considerations

SMTP credentials stored exclusively in environment variables — never in the codebase
Supabase connection credentials managed via env secrets
Input validation at the FastAPI layer using Pydantic before data reaches AI or storage
Gemini API key scoped to the backend only — never exposed to the client

Getting Started

git clone https://github.com/Sameershahh/feedback-form-system
cd feedback-form-system

# Backend
cd api
pip install -r requirements.txt
# Add .env with Supabase, Gmail SMTP, and Gemini credentials
uvicorn main:app --reload

# Frontend
cd ../frontend
npm install
npm run dev

Real-World Use Cases

Conference and workshop organizers collecting post-event feedback
Corporate training teams tracking session quality
SaaS products embedding feedback collection into onboarding flows
Event agencies needing automated client-ready reports

Resources

Curious about the AI prompt engineering behind the Gemini integration, or how the bulk report aggregation works? Drop a comment — happy to go deeper on any part of the stack.

Built by Sameer Shah — AI & Full-Stack Developer | Portfolio

DEV Community