SOPhi: AI SOP Assistant

---

Overview

This project is a powerful AI-powered assistant designed to streamline the creation of detailed and practical Standard Operating Procedures (SOPs). It uses a Retrieval-Augmented Generation (RAG) approach to provide highly contextual and accurate outputs.

By combining an internal knowledge base of your organization's existing SOPs with up-to-date information from the internet, the assistant generates comprehensive documents tailored to your specific needs. This ensures your SOPs are both consistent with your internal policies and current with the latest technology and best practices.

---

Key Features

• Local LLM Integration: Powered by Ollama, the application uses a locally-run, quantized LLM (Mistral) for private, efficient, and cost-free text generation.
• Internal Knowledge Base: Existing SOPs (PDFs, DOCX) are ingested, chunked, and stored in a local ChromaDB vector database. This allows the assistant to pull from your organization's unique policies and procedures.
• Real-time Web Search: The application integrates with SerpAPI to perform real-time web searches, ensuring the generated SOPs include the latest information, security updates, and technical details.
• FastAPI Backend: A robust and scalable RESTful API provides a clean interface for generating SOPs, making it easy to integrate with other applications.

---

Directory Structure

SOPhi/
├── README.md                  - This file
├── .env.example               - Example file for environment variables
├── .gitignore                 - Files to be ignored by Git
├── requirements.txt           - Python dependencies
├── config.py                  - Configuration settings
├── scripts/
│   └── ingest_sops.py         - Script to process and ingest documents
├── src/
│   ├── main.py                - The core FastAPI application
│   ├── llm_service.py         - Handles communication with the local LLM
│   ├── vector_db_service.py   - Manages the ChromaDB vector store
│   ├── web_search_service.py  - Connects to the SerpAPI for web search
├── data/
│   ├── raw_sops/              - **Place your raw SOPs (PDF, DOCX) here**
│   └── embeddings/            - Local vector store data (generated)
└── examples/
    └── example_usage.py       - A script to demonstrate API usage

---

Prerequisites

Before you begin, ensure you have the following installed on your system:

• Python 3.9+
• Ollama (through terminal)
• SerpAPI Key

---

Getting Started

Step 1: Clone the Repository

git clone https://github.com/shussin245/SOPhi.git
cd SOPhi

Step 2: Set Up the Environment

#Create virtual environment
python3 -m venv venv

#Activate virtual environment
source venv/bin/activate  

Step 3: Install Dependencies

pip install -r requirements.txt

Step 4: Configure Environment Variables

Create a .env file in the project root:

SERPAPI_API_KEY="your-serpapi-key-here"

You can edit the .env.example file provided and save as .env.

Step 5: Pull the Models

#Pull the generative LLM
ollama pull mistral

#Pull the embedding model
ollama pull nomic-embed-text

---

Usage Guide

1. Start the Ollama Server

ollama serve

2. Ingest Documents

Add SOPs to data/raw_sops/ and run:

python -m scripts.ingest_sops

This creates the embeddings in data/embeddings/.

3. Run the FastAPI Server

uvicorn src.main:app --reload

Server runs on: http://127.0.0.1:8000

4. Generate an SOP

In a new terminal with your virtual environment activated, run:

python examples/example_usage.py

---

Technical Notes & Best Practices

Performance

• Ingesting SOPs: ~10 minutes
• Generating SOPs: ~20 minutes

Temperature

• 0 → deterministic
• 0.3–0.7 → balanced (recommended)
• 0.9+ → highly creative, risk of hallucinations
  Default: 0.5 for SOPs

Embeddings

• Deterministic vectors for text
• No seed needed for inference
• Seeds only matter for fine-tuning/sampling tasks

Chunking

• Smaller chunks = better semantic precision, worse context
• Larger chunks = better context, worse precision
  Recommended: 500 chars + 100 overlap

---

Next Steps

• Optimize ingestion/generation speed
• Web UI for non-technical users
• Fine-tuned models for domain-specific SOPs

---

Contribution

Pull requests and issues are welcome.