🧠 RAG Corp — Robust PDF & GitHub Retrieval-Augmented Generation System

🚀 A modular, production-grade Retrieval-Augmented Generation (RAG) system with PDF & GitHub ingestion, FAISS vector search, and local LLM inference through Ollama — designed for real-world, offline AI pipelines.

---

🏗️ Overview

RAG Corp is a corporate-style RAG pipeline designed to answer questions from PDF documents and GitHub repositories using local LLMs.
It’s built for offline, secure, and configurable deployments, with robust logging, modular architecture, and full error handling.

---

✨ Features

🗂️ Multi-Source Ingestion

• PDF Loader – Extracts text from uploaded PDFs
• GitHub Repo Loader – Clones repos and indexes documentation/code
• Configurable via YAML – Control chunk size, model name, retriever depth, etc.

🧮 Vector Store with FAISS

• Fast, memory-efficient similarity search
• Persistent offline index in data/index/faiss/
• Incremental updates supported

🤖 Local LLM + Embeddings

• Ollama-powered models (llama3.2:1b, tinyllama:1.1b, deepseek-coder:6.7b)
• Embeddings: nomic-embed-text or fallback to bge-m3
• Fully offline pipeline — no API calls or cloud dependency

🧰 Robust Architecture

• Centralized logging with logging_conf.py
• Error-handled RAG core (graceful model or data fallback)
• Modular structure for easy extension (Chroma, Weaviate, etc.)
• Auto-discovery of configuration paths

💻 Streamlit UI

• Clean, responsive dashboard
• Upload PDFs or Git repos directly
• Real-time logs: document count, index progress, error reports
• Query chat interface for retrieval-based QA

---

🧱 Project Structure

Rag_application/ ├── app/ │ ├── api/ # (Optional) REST endpoints (FastAPI-ready) │ ├── configs/ # YAML configs │ ├── rag-corp/ │ │ ├── rag_core/ │ │ │ ├── embeddings/ # Ollama + HF embedding providers │ │ │ ├── llm/ # Local LLM interface (Ollama) │ │ │ ├── loaders/ # GitHub + PDF data ingestion │ │ │ ├── vectorstores/ # FAISS integration │ │ │ ├── config.py # Auto-path + .env aware settings loader │ │ │ ├── logging_conf.py # Central logging system │ │ │ ├── rag_service.py # RAG logic orchestration │ │ │ └── utils.py # Helpers / validation │ └── ui/ # Streamlit front-end ├── configs/settings.yaml # Global configuration ├── data/ # Local index storage ├── requirements.txt ├── Dockerfile ├── docker-compose.yaml ├── Makefile ├── pyproject.toml └── .env.example

---

⚙️ Tech Stack

Layer	Technology	Purpose
UI	Streamlit	Web-based dashboard
Core Framework	LangChain	RAG orchestration
LLM	Ollama	Local inference engine
Vector Database	FAISS	Embedding similarity search
Config	YAML + dotenv	Dynamic environment setup
Deployment	Docker + Compose	Reproducible environments

---

🧠 How It Works

1. Upload PDF / GitHub URL
2. Text Chunking & Embedding
   • Split into overlapping chunks
   • Generate embeddings (Ollama / HF)
3. Vector Indexing
   • Stored in FAISS index for fast retrieval
4. User Query → Retrieval + LLM Answer
   • Relevant chunks retrieved
   • LLM synthesizes answer with context

---

🧾 Configuration

All runtime settings are stored in configs/settings.yaml:

`yaml env: dev data_dir: data index_dir: data/index/faiss chunk_size: 1000 chunk_overlap: 200 retriever_k: 4 embed_model: bge-m3 llm_model: llama3.2:1b git: branch: main include_exts: [".py", ".md", ".txt"] exclude_dirs: ["pycache", "tests"]

export CONFIG_PATH=./configs/settings.yaml

---

Environment setup:

git clone https://github.com//Rag_application.git cd Rag_application

---

Create environment

python3 -m venv .venv source .venv/bin/activate

Install dependencies

pip install -r requirements.txt

Copy env template

cp .env.example .env

Run

streamlit run app/ui/app.py

Ollama services

ollama serve

ollama pull llama3.2:1b ollama pull tinyllama:1.1b ollama pull deepseek-coder:6.7b-instruct ollama pull nomic-embed-text

UI

---