Sectoral ReAct Agent – Digital Transformation Assistant

📌 Project Overview

This project implements a sector-specific intelligent agent focused on Digital Transformation in Businesses.
The goal is to move beyond static question-answering and build a system that retrieves, reasons, and filters knowledge using a Retrieval-Augmented Generation (RAG) approach.

The agent evaluates whether a user question is within the scope of the sectoral knowledge base and retrieves the most relevant contextual information when applicable.

🎯 Project Goal

• Design a sectoral AI assistant for the field of Digital Transformation
• Use technical documents instead of fixed answers
• Filter out-of-domain questions
• Provide traceable and explainable retrieval results

This project follows the ReAct Agent design philosophy, focusing on reasoning over retrieved information.

🏢 Selected Sector

Digital Transformation in Businesses

Reason for Selection

• Requires strategic, organizational, and technical knowledge
• Highly relevant to modern enterprises
• Suitable for semantic similarity and reasoning-based retrieval

📂 Dataset & Data Source

• Data Type: Plain text document
• File: data/digital_transformation.txt
• Content:
  Analysis of digital transformation impacts on:
  • Business strategy
  • Organizational culture
  • Decision-making processes
  • Competitive advantage
  • Human factor and employee adaptation

The document is manually curated and used as a domain-specific knowledge base.

🧠 Methodology

✅ Method Used: RAG (Retrieval-Augmented Generation)

The system:

1. Splits the document into fixed-size text chunks
2. Converts chunks into vector embeddings
3. Stores embeddings in memory
4. Retrieves the most relevant chunk using cosine similarity
5. Rejects out-of-domain questions using a similarity threshold

🧩 Core Components

Text Chunking

• Chunk size: 400 words
• Ensures semantic coherence

Embedding Model

• sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2
• Supports Turkish language

Similarity Filtering

• Threshold: 0.35
• Prevents hallucination and irrelevant answers

🧪 Benchmark Scenarios

Scenario A – In-Domain Questions

Examples:

• Digital transformation strategy risks
• Organizational resistance
• Impact on competitiveness
• Human factor in transformation

Scenario B – Out-of-Domain Questions

Examples:

• French Revolution
• Quantum computing

Out-of-domain questions are explicitly rejected.

🔍 Output Details

For each user query, the system displays the following information:

• User Question: The original question provided by the user.
• Retrieved Context: The most relevant document chunk retrieved using semantic similarity (if the question is in-scope).
• Similarity Score: A numerical score indicating how closely the question matches the retrieved document content.
• Out-of-Scope Warning: If the similarity score is below the defined threshold, the system notifies that the question is outside the document scope.

📊 Output Image

🤝 Contributing

Contributions are welcome!

📡 Contact

For any queries or collaborations, feel free to reach out!

🌐 GitHub: zeynepcol
👤 LinkedIn: zeynep-col