Add modular support for triage pipeline and documentation

[Mayank2142]

HackerRank Orchestrate: Multi-Domain Support Triage Agent 🤖

Welcome to the Multi-Domain Support Triage Agent, built for the HackerRank Orchestrate 24-Hour Hackathon. This agent is a highly modular, intelligent support triage system designed to process user support tickets across multiple platforms (HackerRank, Claude, Visa), classify them, retrieve context from a domain-specific corpus, and make autonomous decisions to either auto-reply or safely escalate.

✨ Key Features

• Modular Architecture: Clean separation of concerns with dedicated modules for Classification, Retrieval, Decision Making, and Natural Language Generation.
• RAG Integration: Leverages an advanced Retrieval-Augmented Generation (RAG) engine to strictly ground responses in the provided support corpus—minimizing hallucinations.
• Intelligent Routing & Escalation: Autonomously identifies high-risk tickets (e.g., PII requests, billing disputes, critical bugs) and routes them through a strict escalation matrix.
• Traceable Execution: Every decision is backed by a generated justification and logged transparently for seamless audits.

📁 System Architecture

The data processing pipeline is orchestrated through the following components:

• classifier.py: Categorizes incoming tickets (e.g., bug, product_issue, invalid) and extracts targeted product areas.
• retrieval.py: Searches the domain-specific data corpus to fetch grounded evidence for the user query.
• decision.py: Determines whether a ticket can be confidently auto-replied or if it must be escalated to a human agent based on strict risk evaluation.
• generator.py: Crafts professional, empathetic, and evidence-grounded natural language responses.
• pipeline.py: Orchestrates the full lifecycle of a ticket from ingestion to the final output CSV.

---

🚀 Setup

Ensure you have your environment variables set (e.g., OPENAI_API_KEY or ANTHROPIC_API_KEY) using a .env file or exported directly. Then, install the required dependencies:

pip install -r requirements.txt

🛠 Run the Agent

To execute the triage agent on the full support tickets dataset, use the following command:

cd code/
python main.py \
  --tickets ../support_tickets/support_tickets.csv \
  --data ../data/ \
  --output ../support_tickets/output.csv \
  --log ../run_log.txt

🧪 Validation & Testing

To run the agent on the smaller sample dataset for quick testing and validation:

cd code/
python main.py \
  --tickets ../support_tickets/sample_support_tickets.csv \
  --data ../data/ \
  --output ../support_tickets/sample_output.csv \
  --log ../run_log.txt

📊 Output Format

The output is written to a CSV file containing the following strictly formatted columns:
status, product_area, response, justification, request_type

• status: replied or escalated
• product_area: A clear domain category (e.g., screen, fraud, api_integration)
• response: The grounded natural language answer, or a standard escalation message
• justification: A trace of how the agent arrived at its decision
• request_type: product_issue, feature_request, bug, or invalid