Route and analyze customer feedback with Qwen3-VL, Tally, PostgreSQL

Self-Hosted

This workflow provides a complete end-to-end system for capturing, analyzing, and routing customer feedback. By combining local multimodal AI processing with structured data storage, it allows teams to respond to customer needs in real-time without compromising data privacy.

Who is this for?

This is designed for Customer Success Managers, Product Teams, and Community Leads who need to automate the triage of high-volume feedback. It is particularly useful for organizations that handle sensitive customer data and prefer local AI processing over cloud-based API calls.

🛠️ Tech Stack

- Tally.so: For front-end feedback collection. - LM Studio: To host the local AI models (Qwen3-VL). - PostgreSQL: For persistent data storage and reporting. - Discord: For real-time team notifications.

✨ How it works

1. Form Submission: The workflow triggers when a new submission is received from Tally.so. 2. Multimodal Analysis: The OpenAI node (pointing to LM Studio) processes the input using the Qwen3-VL model across three specific layers: - Sentiment Analysis: Evaluates the text to determine if the customer is Positive, Negative, or Neutral. - Zero-Shot Classification: Categorizes the feedback into pre-defined labels based on instructions in the prompt. - Vision Processing: Analyzes any attached images to extract descriptive keywords or identify UI elements mentioned in the feedback. 3. Data Storage: The PostgreSQL node logs the user's details, the original message, and all AI-generated insights. 4. AI-Driven Routing: The same Qwen3-VL model makes the routing decision by evaluating the classification results and determining the appropriate path for the data to follow. 5. Discord Notification: The Discord node sends a formatted message to the corresponding channel, ensuring the support team sees urgent issues while the marketing team sees positive testimonials.