Causal RAG System for Conversational Analysis

Overview

This project implements a Causal Retrieval-Augmented Generation (RAG) system to analyze customer service conversations and produce explainable, evidence-backed answers to analytical queries.

Instead of generating free-form text, the system focuses on:

• Identifying why outcomes occur
• Extracting causal factors
• Grounding explanations in real transcript evidence

---

Project Structure

PredictX/
├── src/
│   └── main.py
├── data/
│   ├── Conversational_Transcript_Dataset.json
│   ├── queries.csv
│   └── submission_output.csv
├── requirements.txt
├── README.md
└── technical_report.pdf

---

How the System Works

1. Retrieve relevant conversations using sentence embeddings
2. Retrieve relevant dialogue turns
3. Tag turns with causal signals (rule-based, deterministic)
4. Aggregate dominant causal factors
5. Produce explainable output with evidence
6. Support follow-up questions using session memory

---

Installation

pip install -r requirements.txt

Running the System

python src/main.py

This will:

1.Read queries from data/queries.csv

2.Run the causal analysis pipeline

3.Write results to data/submission_output.csv

Notes

No external LLM calls are required

All reasoning is transparent and auditable

Designed for analytical and compliance-focused use cases