Customer Support Copilot (Atlan AI Engineer Internship Task)

📌 Overview

This repository contains my submission for the Atlan AI Engineering Internship Challenge.
The project implements a Customer Support Copilot system that can classify tickets, retrieve relevant documentation, and provide an assistant view for resolving queries.

Key Highlights

• Bulk Ticket Classification – Supports multiple tickets at once, with rule-based heuristics as a fallback.
• RAG-backed Assistant – Retrieves answers from ingested documentation; uses LLM generation if an OpenAI API key is available.
• Local Persistence – Chroma DB stores embeddings for efficient retrieval.
• Streamlit UI – Simple interface for interactive exploration.

Dual Modes

• With OpenAI API key → LLM-powered classification and answer generation.
• Without API key → Rule-based classification and local retrieval with Chroma DB.

---

🚀 Project Links & Live Demo

• Live Application: Customer Support Copilot
  Try the working helpdesk application directly in your browser.

• GitHub Repository: Repo Link
  Contains all source code, project files, and documentation.

---

⚙️ Features

• Bulk Ticket Classification: Classifies multiple tickets at once, with fallback heuristics if no OpenAI key is provided.
• RAG-backed Assistant: Uses a retrieval-augmented generation flow to answer queries from ingested docs. If no OpenAI key is set, the assistant still retrieves relevant context locally, but LLM-based answer generation is skipped.
• Chroma DB Support: Stores embeddings locally for efficient retrieval.
• Flexible Setup: Works even without an OpenAI API key (heuristics only).
• Streamlit UI: Simple web app interface for exploration.

---

🚀 Quick Start (Windows PowerShell)

1. Copy .env.example to .env and fill in OPENAI_API_KEY if available.

2. Create and activate a virtual environment, then install requirements:

python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

3. Confirm tickets load:

python ingest.py --list

4. Fetch documentation to data/docs:

python ingest.py --fetch <urls>

5. Launch the Streamlit app:

streamlit run app.py

---

📂 Project Structure

atlan-customer-support-copilot/

├── app.py                    # Streamlit app (UI) & Dashboard
├── ingest.py                 # Handles ticket/doc ingestion
├── classify.py               # Ticket classification logic (topic, sentiment, priority)
├── rag.py                    # RAG helper: answer_query using Chroma + GPT
├── data/
│   ├── chroma_db/            # Local Chroma DB persistence
│   ├── sample_tickets.jsonl  # Example support tickets
│   ├── docs/                 # Optional SDK/docs files for RAG
├── tests/
│   ├── test_ingest.py        # Test ingestion works 
│   ├── test_classify.py      # Test classification logic
├── .gitignore                # Ignore sensitive files
├── render.yaml               # Render deployment config
├── requirements.txt          # Dependencies
├── README.md                 # Project documentation
├── submission.md             # Submission summary

---

🏗️ System Architecture

The following diagram illustrates the complete architecture of the Customer Support Copilot AI System, highlighting the flow from user interaction to AI-driven insights.

---

🛠️ Design Decisions

1. Fallback Heuristics: Even without an OpenAI API key, the app still classifies tickets using rule-based heuristics.

2. Local Persistence: Chroma DB persistence ensures embeddings don’t need to be regenerated every run.

3. Lightweight & Minimal: No heavy dependencies are used outside of requirements.txt, making the app easy to run locally.

4. Dual Mode (API / No API): Designed to run in constrained environments (no external API calls). If an OpenAI key is present, GPT models are used; otherwise, heuristics + local retrieval ensure continuity.

---

🔑 Key Learnings

• Practical use of RAG (Retrieval-Augmented Generation) in a support assistant.
• Trade-offs between heuristic classification and LLM-based classification.
• Best practices for structuring AI apps with a clean, documented repo.

---

✅ Task Completion Status

• ✅ Bulk Ticket Classification – Successfully implemented ticket classification for multiple tickets with topic, sentiment, priority, and confidence scores.
• ✅ RAG-backed Assistant Integration – Implemented a retrieval-augmented generation pipeline using Chroma DB for local persistence, capable of answering tickets/questions based on ingested documentation.
• ✅ Streamlit UI Development – Built an intuitive web interface with separate tabs for bulk ticket dashboard and interactive ticket assistant.
• ✅ Comprehensive Documentation – Provided clear setup instructions, project structure, and usage guidelines in README.md and submission.md.
• Future Enhancements – Potential improvements include refining the UI/UX, integrating advanced classification models, expanding SDK documentation for more accurate RAG responses, and adding automated testing for pipeline validation.

---

📜 License

This project is for internship evaluation purposes only under Atlan’s AI Engineer program. If reusing this work, please ensure proper attribution and follow open-source license guidelines.