aldenirsrv/customer-support-assistant

Customer Support Assistant

Customer Support Assistant is a LoRA fine-tuned version of NousResearch/Llama-2-7b-chat-hf designed to generate customer-support style responses for common ecommerce support requests, such as order cancellation, refunds, invoices, shipping updates, and order tracking.

Model Details

Model Description

This model was fine-tuned using parameter-efficient LoRA/QLoRA techniques on the Bitext customer support chatbot training dataset. The goal is to adapt a Llama 2 chat model to produce structured, polite, customer-service-oriented answers.

• Developed by: Aldenir
• Shared by: aldenirsrv
• Model type: LoRA adapter for causal language modeling
• Language(s): English
• Base model: NousResearch/Llama-2-7b-chat-hf
• Fine-tuning method: LoRA / QLoRA
• Primary use case: Customer support response generation

Uses

Direct Use

This model can be used to generate draft customer-support responses for ecommerce-style support queries, including:

• Order cancellation
• Refund requests
• Shipping address updates
• Order tracking
• Invoice download instructions
• General customer-service assistance

Downstream Use

This adapter can be integrated into:

• Customer-support chatbot prototypes
• Internal support automation experiments
• Synthetic customer-service response generation
• Fine-tuning and evaluation experiments

Out-of-Scope Use

This model should not be used as a fully autonomous production support agent without additional safeguards.

It is not intended for:

• Legal, medical, financial, or safety-critical advice
• Fully automated refund approval or account changes
• Handling sensitive customer data without privacy controls
• Production deployment without human review, monitoring, and policy validation

Bias, Risks, and Limitations

The model may:

• Generate incorrect or generic support instructions
• Produce overly verbose responses
• Reuse placeholders such as {{Order Number}} or {{Customer Support Phone Number}}
• Hallucinate policies, process steps, or company-specific rules
• Fail on queries outside the customer-support domain
• Reflect style limitations from the training dataset

Recommendations

Use this model as a support-response drafting assistant, not as the final authority. For production use, combine it with:

• Retrieval from real company policies
• Tool/API integration for order status and refunds
• Human review for sensitive requests
• Guardrails for privacy, compliance, and escalation
• Evaluation against real customer-support examples

Dataset Coverage

Categories and Intents

The categories and intents covered by the dataset are:

• ORDER

  • cancel_order
  • change_order
  • place_order

• SHIPPING

  • change_shipping_address
  • set_up_shipping_address

• CANCELLATION

  • check_cancellation_fee

• INVOICE

  • check_invoice
  • get_invoice

• PAYMENT

  • check_payment_methods
  • payment_issue

• REFUND

  • check_refund_policy
  • track_refund

Entities (Slot Types)

The dataset includes ~30 entity (slot) types designed to simulate real-world customer-support interactions using structured placeholders.

These entities represent dynamic values that would normally come from real systems (e.g., order IDs, invoices, customer data).

---

🔑 Core Transaction Entities

• {{Order Number}}
  Used for order-related intents such as canceling, modifying, or tracking an order.

• {{Invoice Number}}
  Used in invoice-related queries such as retrieving or checking invoices.

• {{Refund Amount}}, {{Money Amount}}
  Represent monetary values related to refunds or payments.

---

🌐 System & Navigation Entities

• {{Online Order Interaction}}
  Refers to UI actions like “cancel order”, “view order”, etc.

• {{Online Payment Interaction}}
  Represents actions related to payments (e.g., payment methods).

• {{Online Navigation Step}}
  Describes navigation steps in a UI (e.g., “go to settings”).

• {{Online Company Portal Info}}
  Represents login portals or account access pages.

---

📞 Customer Support & Contact Entities

• {{Customer Support Phone Number}}
• {{Customer Support Email}}
• {{Live Chat Support}}
• {{Website URL}}

These simulate support contact channels.

---

👤 User & Account Entities

• {{Client First Name}}, {{Client Last Name}}
• {{Salutation}}
• {{Profile}}, {{Profile Type}}
• {{Account Type}}, {{Account Category}}
• {{Account Change}}, {{Upgrade Account}}

---

⚙️ System Configuration

• {{Settings}}
  Used across multiple intents such as account updates and shipping changes.

---

📦 Delivery & Location Entities

• {{Delivery City}}, {{Delivery Country}}
• {{Store Location}}
• {{Shipping Cut-off Time}}

---

📅 Time-related Entities

• {{Date}}
• {{Date Range}}

---

🎯 Business Context Entities

• {{Program}}
  Used in ordering or subscription-related contexts.

---

These entities are synthetic placeholders:

• The model learns response patterns, not real data
• Outputs may include placeholders directly