Customer Complaint Sentiment Assistant ��

Agent-based emotional classification on customer feedback

Edip Yılmaz

hymmanak@gmail.com

Customer Complaint Sentiment Assistant

Problem statement and proposed solution

Problem

• Customer complaints can vary in tone and urgency. Traditional RPA systems lack emotional intelligence and treat every complaint the same way. This causes inefficient triage and delayed response to high-risk cases.

​

• There is a need for an intelligent system that can analyze the sentiment of customer complaints and take appropriate automated actions depending on emotional tone.

Solution

• The proposed solution is an agent-based workflow that reads customer complaints from a spreadsheet and analyzes their sentiment using a predefined classification logic. Based on the result (positive, neutral, or negative), different actions can be taken automatically or flagged for human intervention.

​

• The system includes:
• - Retry and exception handling
• - History check to avoid duplicates
• - SentimentAgent for tone evaluation
• - Modular, reusable XAML components

3

Benefits and technologies used

4

​

Benefits, impact and outcomes

• Enables fast classification of customer complaints based on emotional toneFaster optimization cycles through automated code generation
• Reduces time spent on manual review of customer feedbackKnowledge democratization making ML optimization accessible to engineers of all levels
• Flags high-risk complaints for escalation, improving customer retentionPotential for substantial ROI through recovered engineering productivity
• Avoids duplicate processing with history tracking
• Provides a scalable foundation to integrate with LLM-based APIs (e.g., GPT)
• Modular design makes it easy to customize or extend the agent for different industries

​

​

Details

Solution architecture

5

6

Project Main ScreenShot

7

8

Demo

9

Note: Since the agent currently runs in backend without a UI, I extracted the output from logs and displayed it here. These are real sentiment responses from real test complaints, showing how the agent would act in production.

English Version of Agent Reply to Customer:

10

2025-07-06 17:47:51.149 GMT+3: Information: ---

Tone: Angry

Category: Payment Issue

Priority: High

Explanation: The customer is angry due to a 3-hour flight delay, lack of communication, and missed work. They are also requesting a refund and have provided sensitive financial information, which cannot be processed directly.

---

Urgent Help Case: The customer experienced a 3-hour flight delay without ana prior communication, resulting in missed work. They are requesting a refund and have shared their IBAN details. Immediate attention is required to address their concerns and provide a resolution.

Company Reply: "Sayın Mert Yılmaz, yaşadığınız gecikme ve iletişim eksikliği nedeniyle içtenlikle özür dileriz. Talebinizi öncelikli olarak ele alacağız. IBAN bilgilerinizi güvenliğiniz için lütfen bu platformda paylaşmayınız. Ekibimiz en kısa sürede sizinle iletişime geçerek çözüm sunacaktır. Anlayışınız için teşekkür ederiz."

"Sayın Mert Yılmaz, yaşadığınız gecikme ve iletişim eksikliği nedeniyle içtenlikle özür dileriz. Talebinizi öncelikli olarak ele alacağız. IBAN bilgilerinizi güvenliğiniz için lütfen bu platformda paylaşmayınız. Ekibimiz en kısa sürede sizinle iletişime geçerek çözüm sunacaktır. Anlayışınız için teşekkür ederiz.«

​

English Translation:�"Dear Mert Yılmaz, we sincerely apologize for the delay and lack of communication you experienced. We will treat your request as a priority. For your safety, please do not share your IBAN details on this platform. Our team will contact you as soon as possible to provide a resolution. Thank you for your understanding."

Thank you for watching this humble MVP.�It’s far from perfect, but it's real — and it’s built with intent, clarity, and potential.