🎓 AI Engineering Knowledge Assistant

AI Engineering Knowledge Assistant (Hybrid RAG)

An LLM-powered internal AI assistant built using LangChain, OpenAI GPT-4o-mini, and ChromaDB to answer technical questions from documentation using Retrieval-Augmented Generation (RAG).

This project simulates a real-world enterprise AI tool used for:

• AI assistants and chatbots
• Knowledge base automation
• Engineering documentation Q&A
• LLM-driven internal tools for business and technical teams

---

Overview

The AI Engineering Knowledge Assistant is a CLI-based Hybrid RAG pipeline that retrieves relevant documents from a local vector database (ChromaDB) and generates context-grounded answers using an LLM (GPT-4o-mini).

Unlike basic chatbots, this system:

• Uses real project files as a knowledge base
• Prevents hallucination with strict context grounding
• Tracks cost, performance, and confidence metrics
• Implements production features such as retry logic, caching, and fallback

---

Key Features

🔍 Retrieval System

• Similarity Search (semantic retrieval)
• MMR Retrieval (diverse document selection)
• Dynamic Retriever Switching
• Source File Filtering (enterprise-style control)

🤖 Production AI Capabilities

• Hybrid RAG Architecture (Local Embeddings + LLM)
• Modern LCEL Runnable Chain (LangChain 2024+)
• Retry Logic with Exponential Backoff
• Safe Fallback (zero extra LLM cost)
• Query Caching System
• Confidence Scoring
• Cost Tracking per Query

📊 Observability & Analytics

• Response time tracking
• Token usage monitoring
• API cost tracking
• ROI analytics
• Export to JSON/CSV for analysis

---

🏗️ System Architecture (Production Hybrid RAG)

User Query (CLI - main.py)
↓
EnhancedProductionRAG (Core Engine)
↓
Query Processing Layer
├── Cache Check (Deterministic Caching)
├── Retry Logic (Exponential Backoff)
└── Metrics Initialization
↓
Dynamic Retrieval Layer
├── Similarity Retriever (Fast Semantic Search)
└── MMR Retriever (Diverse Document Retrieval)
↓
Chroma Vector Database (Local)
↓
HuggingFace Embeddings (Sentence Transformers)
↓
LCEL Runnable Chain (LangChain 2024+)
├── Context Formatting
├── Prompt Template (Anti-Hallucination)
└── GPT-4o-mini (LLM Generation)
↓
Post-Processing Layer
├── Confidence Scoring
├── Cost Tracking (OpenAI Callback)
├── Source Attribution
└── Safe Fallback (Context-Based Response)
↓
Final Output
Answer + Sources + Metrics + Cost + Confidence

---

🚀 Demo (CLI - Hybrid RAG in Action)

---

🔄 End-to-End RAG Pipeline Flow

1. User asks a technical question via CLI
2. System checks cache for repeated queries
3. Dynamic retriever fetches top-k relevant documents
4. Documents are filtered and formatted as context
5. LCEL chain constructs prompt with strict grounding rules
6. GPT-4o-mini generates a context-aware answer
7. System calculates confidence, cost, and performance metrics
8. Final response is returned with sources and analytics

---

Project Structure

AI-Engineering-Knowledge-Assistant/
│
├── .github/
│ └── workflows/
│ └── ci.yml # GitHub Actions CI/CD pipeline
│
├── assets/ # Demo screenshots (README images)
│ ├── demo_1.png
│ ├── demo_1.5.png
│ ├── demo_2.png
│ └── demo_2.2.png
│
├── data/
│ ├── documents/ # Knowledge base files (PDF, MD, TXT, code)
│ ├── vector_store/ # Chroma vector DB (auto-generated)
│ └── query_cache.pkl # Cached query results (auto-created)
│
├── exports/ # Exported analytics (JSON/CSV)
│
├── src/
│ ├── ingestion/
│ │ ├── init.py
│ │ └── knowledge_base.py # Document loading + embeddings + vector DB
│ │
│ ├── rag/
│ │ ├── init.py
│ │ └── rag_pipeline.py # Core EnhancedProductionRAG (main engine)
│ │
│ └── interface/
│ ├── init.py
│ └── main.py # CLI entry point (interactive assistant)
│
├── .env # Environment variables (NOT pushed to GitHub)
├── .gitignore # Ignore secrets, cache, vector DB
├── requirements.txt # Project dependencies
├── README.md # Project documentation
└── LICENSE # (Optional but recommended)

---

Tech Stack

Category	Technology
Language	Python 3.10+
LLM	OpenAI GPT-4o-mini
Framework	LangChain (LCEL Architecture)
Vector Database	ChromaDB (Local)
Embeddings	HuggingFace Sentence Transformers
Interface	CLI (Production Stable)
CI/CD	GitHub Actions

---

Installation

1. Clone the Repository

git clone https://github.com/Hanah7511/AI-Engineering-Knowledge-Assistant.git
cd AI-Engineering-Knowledge-Assistant

### 2. Create Virtual Environment
python -m venv venv
# Windows
venv\Scripts\activate
# Mac/Linux
source venv/bin/activate  

### 3. Install Dependencies

pip install -r requirements.txt

### 4. Add Environment Variables

Create a .env file in the root directory:
OPENAI_API_KEY=your_api_key_here

---

## CI/CD Pipeline

This project includes a GitHub Actions CI pipeline that:
-Installs dependencies automatically
-Validates RAG pipeline imports
-Detects missing packages and runtime errors
-Ensures code stability on every push

Workflow location:
.github/workflows/ci.yml

---

## Production Design Highlights

- Modular architecture (ingestion, retrieval, generation separation)
- Laptop-safe optimization (low memory footprint)
- Fail-safe fallback when API fails
- Deterministic caching for repeated queries
- Enterprise-style source filtering and observability

---
## Future Improvements

-FastAPI deployment (REST API)
-Docker containerization
-Hybrid Search (BM25 + Vector)
-RAG evaluation (RAGAS)
-Web UI (optional)

---

## Author

Hana Al Haris
Final Year AI/ML Student
Portfolio Project – AI Engineering & RAG Systems