Specpilot

This repository aims to provide the tools to deploy an AI-powered assistant for 3GPP specification, combining two main components:

1. 3GPP Specification Extraction (Python Script):

   • Downloads official 3GPP specifications
   • Extracts and converts them to markdown format for easy processing and search
   • Output is organized by release in the spec_md/ directory

2. ChatGPT-like Features (Ollama + AnythingLLM):

   • Runs Ollama locally to provide LLM inference (e.g., qwen3:30b-a3b, deepseek-r1)
   • Deploys AnythingLLM via Docker to enable a chat interface over your documents
   • Allows advanced querying, context window management

---

Features

• Automated download and conversion of configured 3GPP specifications to markdown
• Local LLM inference using Ollama
• ChatGPT-like interface for document Q&A via AnythingLLM

---

Installation & Setup

Prerequisites

• Docker
• Docker Compose
• Python 3.8+
• Ollama (for local LLMs)

Steps

1. Clone this repository:

   git clone https://github.com/herlesupreeth/specpilot.git
   cd specpilot
   mkdir -p llm_storage
   

2. Install Ollama and Download Models:

   curl -fsSL https://ollama.com/install.sh | sh
   ollama pull qwen3:30b-a3b
   ollama pull nomic-embed-text
   # You can use other Ollama models as well. See https://ollama.com/library for available models.
   

---

3GPP Specification Extraction (Python Script)

Currently, the repository contains 38 series specifications from Release 15-18 already downloaded and converted to markdown.

If you want to download additional specifications modify below two parameters in 3gpp_extraction.py:

# Add more releases as needed
RELEASES = ["Rel-15", "Rel-16", "Rel-17", "Rel-18"]
# Add more series as needed
SERIES = ["38_series"]

Then run:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python3 3gpp_extraction.py

This will download, extract, and convert 3GPP specs to markdown in spec_md/.

---

Deployment of ChatGPT-like Features (Ollama + AnythingLLM)

1. (Optional) Configure environment variables in .env as needed.

2. Run Ollama:

   Note: If Ollama is already running, you can skip this step.

   ollama serve qwen3:30b-a3b
   # Replace 'qwen3:30b-a3b' with any other model you have pulled if desired.
   

   Visit http://localhost:11434 to verify Ollama is running.

3. Ensure Docker and Docker Compose are installed.

4. Start AnythingLLM:

   source .env
   docker-compose up -d
   

5. Access the UI at http://localhost:3001.

---

Directory Structure

• llm_storage/ - Persistent storage for models, documents, and context
• spec_md/ - 3GPP specification markdown files
• 3gpp_extraction.py - Extraction and processing script
• docker-compose.yml - Deployment configuration

---

Contributing

Pull requests and issues are welcome! Please see the LICENSE for details.

License

BSD 2-Clause License

---

Troubleshooting

Common Issues

• Ollama not running:
  Ensure Ollama is installed and the model is served. Run ollama serve <model> and check http://localhost:11434.

• Ollama model not found:
  Make sure you have pulled the model using ollama pull <model_name>. Check available models with ollama list.

• Ollama not accessible:
  If Ollama is running but not accessible, check if the port (default 11434) is blocked by a firewall or if Ollama is configured to use a different port. To allow access in firewall run sudo ufw allow 11434.

• Docker Compose errors:
  Verify Docker and Docker Compose are installed and running. Use docker --version and docker-compose --version to check.

• Missing Python dependencies:
  Run pip install -r requirements.txt to install required packages.

• Specs not downloading:
  Check your internet connection and verify the RELEASES and SERIES parameters in 3gpp_extraction.py.

• UI not accessible:
  Ensure Docker is running and the container is up. Check logs with docker-compose logs for any errors. Ensure SERVER_PORT in .env is open and not blocked by firewall. To allow access in firewall run sudo ufw allow ${SERVER_PORT}.

• Port conflicts:
  If the specified ports are already in use, change the SERVER_PORT in .env and redeploy.

If encounter other issues, please open an issue in the repository with error details.