Automatically analyze customer feedback using AI and route it to the right team channels — eliminating manual triage and enabling real-time sentiment tracking.
- Manual reading and categorization of every customer submission
- Slow response times due to human triage bottlenecks
- Prone to misrouting and classification errors
- No systematic tracking of sentiment or recurring issues
- Image attachments require additional manual review
- Valuable customer insights lost in cluttered inboxes
An end-to-end intelligent automation that:
- Captures customer feedback via public web form
- Analyzes sentiment automatically (Positive, Negative, Neutral)
- Classifies feedback into relevant categories without manual tagging
- Extracts keywords from uploaded images using vision capabilities
- Stores all processed data in a database for tracking and analytics
- Routes feedback to the appropriate team channel using smart decisions
- Sends rich notifications with complete feedback details and insights
💡 Once deployed, every submission is automatically analyzed, categorized, stored, and routed — with zero manual intervention.
For a full walkthrough of the system, watch the complete demo video.
- 📝 Public Feedback Collection — Collect customer feedback via shareable Tally.so forms (Name, Email, Feedback, optional Image)
- ⚡ Instant Webhook Processing — Submissions trigger the workflow in real-time with zero delays
- 🧠 AI-Powered Sentiment Analysis — Automatically detect customer sentiment (Positive, Negative, Neutral) using local LLM
- 🏷️ Zero-Shot Text Classification — Categorize feedback into pre-defined labels without model training
- 🖼️ Multimodal Image Analysis — Extract keywords from uploaded images using Qwen2-VL vision capabilities
- 🔀 Intelligent Channel Routing — AI-driven decisions to route feedback to appropriate Discord channels
- 💾 Automated Data Storage — All processed feedback stored in PostgreSQL with structured schema
- 💬 Real-Time Discord Alerts — Instant embed notifications with feedback details and AI insights
- 🏠 Fully Local AI Processing — Privacy-first approach with on-device LLM inference via LM Studio
- 🐳 Dockerized Deployment — One-command setup using Docker Compose for PostgreSQL, Adminer, and n8n
- 💸 100% Free & Open Source — Built entirely on free-tier services and open-source tools with no ongoing costs
| Category | Technology | Description | Link |
|---|---|---|---|
| Workflow Automation | n8n Community Edition | Self-hosted workflow orchestration for form processing, AI analysis, and routing | n8n.io |
| Form Handling | Tally.so | Free form builder for public feedback collection (Name, Email, Feedback, optional Image) | tally.so |
| AI/ML Inference | OpenAI Node + LM Studio + Qwen2-VL-7B-Instruct | n8n's OpenAI node configured with custom Base URL pointing to LM Studio's local server hosting Qwen2-VL (GGUF, Q4_K_M). Tasks: • Sentiment analysis • Zero-shot text classification • Image keyword extraction • Intelligent routing decisions |
OpenAI Node / LM Studio / Qwen2-VL |
| Database | PostgreSQL + Adminer | Relational database for storing processed feedback with lightweight management UI | PostgreSQL / Adminer |
| Notifications | Discord | Channel-based routing for customer feedback to relevant team channels | discord.com |
| Containerization | Docker + Docker Compose | Multi-container orchestration for PostgreSQL, Adminer, and n8n via docker-compose.yml |
Docker / Compose |
User submits feedback form (Tally.so)
↓
Tally Trigger
↓
Field Mapping
↓
┌─────────────────────────────┼─────────────────────────────┐
↓ ↓ ↓
Sentiment Text Fetch Image
Analysis Classification ↓
↓ ↓ ┌────────┴────────┐
Sentiment Classification ↓ ↓
LLM LLM Has Image? No Image?
↓ ↓ ↓ ↓
Sentiment Classification Image Empty
Parser Parser Keyword Keywords
↓ ↓ Extraction Handler
│ │ ↓ ↓
│ │ └────────┬────────┘
│ │ ↓
│ │ Image Results Merge
│ │ ↓
└─────────────────────────────┴─────────────────────────────┘
↓
AI Results Merge
↓
Data Aggregation
↓
Save to PostgreSQL
↓
Decision Logic
↓
Routing LLM
↓
Route Parser
↓
Channel Router
↓
┌───────────────────────────┼───────────────────────────┐
↓ ↓ ↓
#general-inquiries #happy-customers #support-urgent
↓ ↓ ↓
└───────────────────────────┴───────────────────────────┘
↓
Build Discord Message
↓
Format Embed Data
↓
Send Discord Notification
Before you begin, make sure you have the following set up:
| Software | Version | Purpose | Installation Guide |
|---|---|---|---|
| Docker | 28.5.1+ | Container runtime for services | Install Docker |
| n8n | 1.121.3 (latest) | Workflow automation platform | n8n Docker Setup |
| LM Studio | 0.3.36+ | Local LLM hosting with OpenAI-compatible API | Download LM Studio |
Verify Installation:
# Check Docker
docker --version
# Expected: Docker version 28.0.0 or higher| Requirement | Details |
|---|---|
| Model | Qwen2-VL-7B-Instruct |
| Format | GGUF |
| Quantization | Q4_K_M |
| Server Port | 1234 (default) |
| Base URL | http://192.168.56.1:1234/v1 |
Setup Steps:
- Download and install LM Studio
- Download
Qwen2-VL-7B-Instructmodel (GGUF, Q4_K_M quantization) - Load the model in LM Studio
- Start the local server (default port: 1234)
- Note your machine's IP address for Base URL configuration
📘 Guide: LM Studio Local Server Setup
| Service | Free Tier | Purpose | Signup / Setup |
|---|---|---|---|
| Tally.so | ✅ Yes | Form builder for feedback collection | Create Account |
| Discord | ✅ Yes | Team notifications & channel routing | Create Server |
| PostgreSQL | ✅ Yes (Docker) | Database for storing processed feedback | Included via Docker Compose |
| Adminer | ✅ Yes (Docker) | Database management UI | Included via Docker Compose |
Ensure you have the following credentials ready before configuring n8n:
- API Key (for webhook authentication)
- Allowed HTTP Request Domains:
All
📘 Guide: Getting your Tally API Key
- API Key (can be any string, e.g.,
lm-studioor dummy key) - Base URL:
http://<YOUR_IP>:1234/v1 - Organization ID: (optional, leave blank)
- Allowed HTTP Request Domains:
All
⚠️ Note: Since you're using LM Studio locally, the API Key can be any placeholder value. The Base URL must point to your LM Studio server.
- Bot Token (
MTI...) - Allowed HTTP Request Domains:
All - Bot Permissions:
Send Messages,Embed Links - Bot added to your Discord server
📘 Guide: Creating a Discord Bot
- Host:
db(Docker service name) - Database:
postgres - User:
<user_name> - Password:
<password> - Port:
5432(default)
📘 Note: These credentials are configured in your
docker-compose.ymlfile.
Before running the workflow, ensure the following are created:
- Create a form with the following fields:
| Field Name | Type | Required |
|---|---|---|
| Name | Text | ✅ Yes |
| ✅ Yes | ||
| Feedback | Long Text | ✅ Yes |
| Image | File Upload | ❌ No |
- Publish the form and copy the Form ID from the URL
- Configure webhook to point to your n8n instance
- Create a Discord server (or use existing)
- Create the following channels:
| Channel Name | Purpose |
|---|---|
#general-inquiries |
General feedback & questions |
#happy-customers |
Positive feedback & testimonials |
#support-urgent |
Urgent issues & negative feedback |
- Note the Channel IDs for each channel
- Add your Discord Bot to the server
💡 Tip: To get Channel ID, enable Developer Mode in Discord settings, then right-click channel → Copy ID
- Database and table will be auto-created via Docker Compose
- Verify connection using Adminer at
http://localhost:8080
| Service | Local URL | Port | Access Method |
|---|---|---|---|
| n8n | http://localhost:5678 |
5678 | Browser |
| Adminer | http://localhost:8080 |
8080 | Browser (PostgreSQL management) |
| LM Studio | http://192.168.56.1:1234/v1 |
1234 | API only |
⚠️ Important: PostgreSQL runs on port5432but is not browser-accessible. Use Adminer atlocalhost:8080to manage your database.
# Clone the repo
git clone https://github.com/Neloy-Barman/AI-Powered-Customer-Feedback-Automation.git
# Navigate to the project directory
cd AI-Powered-Customer-Feedback-AutomationBefore starting the containers, configure your PostgreSQL credentials in docker-compose.yml:
# Open docker-compose.yml and update these values
services:
db:
environment:
POSTGRES_USER: <user_name> # Preferred username
POSTGRES_PASSWORD: <password> # Set a secure password
POSTGRES_DB: postgres # Database Name
⚠️ Important: Remember these credentials — you'll need them when configuring n8n PostgreSQL connection.
Ensure Docker Desktop is running in the background, then start the containers:
# Start the containers in detached mode
docker compose up -dThis will spin up:
| Service | URL | Purpose |
|---|---|---|
| n8n | http://localhost:5678 |
Workflow automation |
| Adminer | http://localhost:8080 |
Database management UI |
| PostgreSQL | Port 5432 |
Data storage |
💡 Check container logs in Docker Desktop GUI to confirm successful startup.
Verify PostgreSQL is Running:
Access Adminer at http://localhost:8080 and login with:
| Field | Value |
|---|---|
| System | PostgreSQL |
| Server | db |
| Username | <user_name> (or your configured username) |
| Password | <password> (your configured password) |
| Database | postgres (or your configured database) |
- Open LM Studio application
- Load
Qwen2-VL-7B-Instructmodel (GGUF, Q4_K_M) - Navigate to Local Server tab
- Click Start Server (default port:
1234) - Note your machine's IP address (e.g.,
192.168.56.1)
⚠️ Ensure the server is running before activating the n8n workflow.
Open your browser and navigate to:
http://localhost:5678
Create an account or log in if prompted.
- Click the
...menu in the top-right corner - Select Import from File
- Choose
AI-Powered-Customer-Feedback-Automation.jsonfrom theworkflowsdirectory
Set up the required credentials in n8n:
| Credential | Navigation | Required Values |
|---|---|---|
| Tally.so | Settings → Credentials → Add → Tally | API Key |
| OpenAI | Settings → Credentials → Add → OpenAI | API Key (any placeholder), Base URL (http://<YOUR_IP>:1234/v1) |
| PostgreSQL | Settings → Credentials → Add → Postgres | Host: db, Database: postgres, User, Password |
| Discord | Settings → Credentials → Add → Discord Bot | Bot Token |
📘 Refer to Prerequisites for detailed credential setup guides.
Create the following channels in your Discord server:
| Channel Name | Purpose |
|---|---|
#general-inquiries |
General feedback & questions |
#happy-customers |
Positive feedback & testimonials |
#support-urgent |
Urgent issues & negative feedback |
Then update the workflow nodes with corresponding Channel IDs.
💡 Tip: Enable Developer Mode in Discord (Settings → Advanced), then right-click channel → Copy ID
Create a public form with the following fields:
| Field Name | Type | Required |
|---|---|---|
| Name | Text | ✅ Yes |
| ✅ Yes | ||
| Feedback | Long Text | ✅ Yes |
| Image | File Upload | ❌ No |
Publish the form and connect it to the workflow trigger.
Open the imported workflow and update these nodes:
| Node | Parameter | Value |
|---|---|---|
| Tally Trigger | Form | Select your Tally form |
| Channel Router | Channel IDs | Your Discord channel IDs |
| Save to PostgreSQL | Credentials | Your PostgreSQL credentials |
- Toggle the Active switch in the top-right corner
- Status should show Active (green)
Your automation is now live. Submit a test feedback through your Tally.so form and watch the magic happen! ✨
💡 Head over to 📘 Usage to learn how to test and manage the workflow.
| User Type | Entry Point | Purpose |
|---|---|---|
| Customer | Tally.so Form URL | Submit feedback with optional image |
| Admin | n8n Interface (localhost:5678) |
Monitor and manage workflow |
| Admin | Adminer (localhost:8080) |
View and manage database records |
| Team | Discord Channels | Receive routed feedback notifications |
Share your Tally.so form URL publicly. Customers fill in the following:
| Field | Type | Required | Example |
|---|---|---|---|
| Name | Text | ✅ Yes | Jane Smith |
| ✅ Yes | [email protected] |
||
| Feedback | Long Text | ✅ Yes | The product quality exceeded my expectations! |
| Image | File Upload | ❌ No | damaged_product.jpg |
💡 The Feedback content determines sentiment, classification, and channel routing automatically.
To verify everything works correctly:
- Ensure LM Studio server is running with model loaded
- Open n8n interface at
http://localhost:5678 - Open the AI-Powered-Customer-Feedback-Automation workflow
- Click on each node sequentially
- Click Test Step to execute individually
- Verify outputs at each stage
Sample Test Data:
| Scenario | Name | Feedback | Expected Channel | |
|---|---|---|---|---|
| Positive | Jane Smith | [email protected] | The product quality exceeded my expectations! | #happy-customers |
| Negative | John Doe | [email protected] | I got a broken Iphone delivered at my home. | #support-urgent |
| Neutral | Alex Brown | [email protected] | Are you guys available at 12:00 pm? | #general-inquiries |
After a successful submission:
| Output | Location | What to Expect |
|---|---|---|
| Database Record | PostgreSQL (via Adminer) | New row with feedback and AI analysis |
| Discord Message | Routed channel | Embed notification with feedback details |
Database Record Structure:
| Column | Example Value |
|---|---|
| id | 1 |
| name | John Doe |
[email protected] |
|
| feedback | I got a broken Iphone delivered at my home. |
| sentiment | Negative |
| category | Product Quality & Features |
| image_keywords | broken, cracked, or damaged product |
| img_url | https://tally.so/uploads/image123.jpg |
Discord Notifications:
Positive Feedback → #happy-customers
🎉 Happy Customer Alert
Positive feedback from a satisfied customer.
👤 Customer Info
Jane Smith ([email protected])
✉️ Feedback
The product quality exceeded my expectations!
📝 Model Analysis
Customer Jane Smith praises our service, noting that it exceeded their expectations.
📝 Automated Feedback Summary
Neutral Feedback → #general-inquiries
🔹 General Feedback Alert
A new general feedback or inquiry has been received.
👤 Customer Info
Alex Brown ([email protected])
✉️ Feedback
Are you guys available at 12:00 pm?
📝 Model Analysis
Customer Alex Brown inquires about availability at 12:00 pm.
📝 Automated Feedback Summary
Negative Feedback → #support-urgent
⚠️ Urgent Support Required
Immediate attention needed for this customer feedback.
👤 Customer Info
John Doe ([email protected])
✉️ Feedback
I got a broken Iphone delivered at my home.
📝 Model Analysis
Customer John Doe reports a broken Iphone delivered at their home, requiring immediate resolution to prevent further inconvenience.
📝 Automated Feedback Summary
Each submission undergoes the following analysis:
Sentiment Analysis:
| Output | Description |
|---|---|
Positive |
Satisfied customer, praise, recommendations |
Negative |
Complaints, issues, urgent problems |
Neutral |
General questions, inquiries |
Text Classification Categories:
| Category | Description |
|---|---|
Product Quality & Features |
Comments about product performance, durability, functionality, or characteristics |
Shipping & Delivery |
Issues or feedback about shipping speed, packaging, delivery experience, or logistics |
Customer Support Experience |
Interactions with support team, response times, helpfulness of staff |
Pricing & Billing |
Comments about prices, charges, payment issues, value for money |
Return & Refund Request |
Requests to return products or get refunds, return policy questions |
Website / App Usability |
Navigation issues, checkout problems, app bugs, user interface feedback |
Positive Feedback |
General satisfaction, compliments, recommendations, praise |
General Inquiry |
Questions or comments that don't fit other categories |
Image Keyword Extraction:
| Keyword | Description |
|---|---|
broken, cracked, or damaged product |
Product defects or physical damage |
damaged or torn packaging |
Packaging issues during shipping |
showing wrong item delivery |
Incorrect item received |
new and intact product |
Product in good condition |
a screenshot of a website or order confirmation |
Digital proof or reference |
a blurry, dark, or unclear photo |
Poor image quality |
an irrelevant photo (e.g., a person, pet, or scenery) |
Non-related image content |
Channel Routing Logic:
| Sentiment | Routed To | Alert Type |
|---|---|---|
| Positive | #happy-customers |
🎉 Happy Customer Alert |
| Negative | #support-urgent |
|
| Neutral | #general-inquiries |
🔹 General Feedback Alert |
✅ Submission successful when:
- New record appears in PostgreSQL database
- Discord notification received in correct channel
- Sentiment correctly identified
- Category properly classified
- Image keywords extracted (if image provided)
- Model analysis summary generated
❌ Something went wrong if:
- No new record in database
- No Discord notification
- Error shown in n8n execution log
- LM Studio server not responding
Troubleshooting:
| Issue | Possible Cause | Solution |
|---|---|---|
| No AI analysis | LM Studio not running | Start LM Studio server |
| Database error | PostgreSQL container down | Run docker compose up -d |
| No Discord message | Invalid Bot Token | Verify Discord credentials |
| Webhook not triggered | Form not connected | Check Tally.so integration |
| Wrong channel routing | Sentiment misclassified | Check LLM model is loaded correctly |
Check Workflow Status:
- Open n8n interface at
http://localhost:5678 - Locate AI Customer Feedback Automation workflow
- Check the toggle — Green = Active, Gray = Inactive
Stop the Workflow:
docker compose downRestart the Workflow:
docker compose up -dView Logs:
docker compose logs -fAccess Database:
- Open Adminer at
http://localhost:8080 - Login with PostgreSQL credentials
- Navigate to feedback table to view all records
Use Adminer to run queries for insights:
-- Count feedback by sentiment
SELECT sentiment, COUNT(*) FROM feedback GROUP BY sentiment;
-- Recent negative feedback requiring attention
SELECT * FROM feedback WHERE sentiment = 'Negative' ORDER BY id DESC LIMIT 10;
-- Most common categories
SELECT category, COUNT(*) FROM feedback GROUP BY category ORDER BY COUNT(*) DESC;
-- All feedback with images attached
SELECT * FROM feedback WHERE img_url IS NOT NULL;This project is licensed under the MIT License — see the LICENSE file for details.
Neloy Barman – AI Agent Builder | Automation Engineer | Chatbot Developer
- 💼 LinkedIn: Neloy Barman
- 🐙 GitHub: Neloy-Barman
- ✍️ Medium: @neloyupwork018
- 🌐 Portfolio: neloy-barman.github.io
- 📧 Email: [email protected]
Built with ☕ and automation in mind by Neloy Barman
📝 Documentation crafted with assistance from Claude AI
