Industrial Knowledge Base AI Assistant Using RAG and Qwen2.5

Project Overview

This project involved development of a local industrial AI assistant for Iran Alloy Steel Company using the company’s internal technical wiki as the knowledge source.

The goal was to make internal engineering and operational knowledge easier to access through a conversational AI interface rather than manual wiki browsing.

The system used a Retrieval-Augmented Generation architecture with a locally deployed Qwen2.5 language model to answer technical questions using retrieved company documentation.

My Role

My responsibilities included:

• local LLM deployment
• RAG pipeline design
• wiki content processing
• document chunking and indexing
• semantic retrieval integration
• AI assistant workflow development
• industrial knowledge-base system design

I worked on turning the company’s technical wiki into a searchable and conversational knowledge system.

RAG Architecture

The system used a Retrieval-Augmented Generation pipeline.

The workflow included:

• collecting internal wiki documents
• cleaning and structuring technical content
• splitting documents into retrievable chunks
• generating embeddings for semantic search
• retrieving relevant context for each user query
• generating grounded responses using Qwen2.5

This allowed the assistant to answer questions based on internal technical documentation rather than relying only on the model’s general knowledge.

Local LLM Deployment

A local Qwen2.5 model was deployed to support private industrial AI usage.

This was important because the company’s technical wiki contained internal engineering knowledge that should not be sent to public cloud AI services.

The local deployment enabled:

• privacy-preserving inference
• internal network operation
• conversational technical support
• company-specific knowledge retrieval

Industrial Knowledge Management

The assistant was designed to help technical users quickly access information from the company’s knowledge base.

Instead of searching through wiki pages manually, users could ask engineering questions and receive responses grounded in retrieved documentation.

This made the system useful for:

• technical troubleshooting
• operational knowledge lookup
• engineering documentation search
• onboarding and knowledge transfer

Engineering Challenges

The project required solving several practical AI-system challenges:

• preparing unstructured wiki content for retrieval
• improving semantic search over industrial terminology
• reducing hallucination by grounding answers in source documents
• designing prompts for technical answer generation
• deploying LLM inference locally rather than through external APIs

Engineering Impact

This project demonstrated practical deployment of local LLM systems for industrial knowledge management.

It strengthened my experience in:

• enterprise AI
• local LLM deployment
• RAG systems
• semantic retrieval
• industrial AI assistants
• technical knowledge management

The project also connects directly to my broader work in robotics, industrial automation, and AI systems by showing how local LLMs can support engineering teams and technical operations.