🤖 Project Erasmus — Reasoning Engine

Project Erasmus is a local, containerized reasoning engine designed to analyze software system data, visualize relationships, and generate GenAI-powered insights. Inspired by Erasmus from Dune, it explores how local AI can reason about real-world software risks.

---

🚀 Features

• Parses SBOMs generated by Syft CLI (CycloneDX JSON)
• Runs Grype CLI scans to detect CVEs
• Joins in metadata from vendor_geo.csv to add vendor HQ + region context
• Visualizes data and vulnerability trends in a Streamlit dashboard
• Integrates Ollama with a dropdown to switch between 3 local GenAI models for contextual summaries and Q&A
• Fully Dockerized — portable, offline, and reproducible

---

🧠 Why Local LLMs?

Erasmus lets you query your data entirely offline using:

• qwen2.5-coder:7b → for technical/structured SBOM interpretation
• gemma:2b → for lightweight analysis, fast on M1/M2 chips
• openhermes:latest → for broader contextual or general summaries

Each model serves a different thinking style. You can switch between them using the Streamlit UI dropdown.

---

✨ Optional Enhancements (paid or enterprise)

For users who want to go deeper:

• Append data from commercial APIs like IPinfo, WHOISXML, or Crunchbase to auto-fill vendor HQ, size, and sector
• Add package popularity or maturity ratings using OSS Index or Libraries.io
• Feed results into a GRC tool for automated software supplier risk scoring

---

🏗 Tech stack

• Syft CLI → SBOM generation
• Grype CLI → vulnerability scanning
• Python + pandas → metadata joining + preprocessing
• Streamlit → visual dashboard with interactive LLM input
• Ollama → GenAI summarization + Q&A using local models
• Docker + Compose → fully containerized environment

---

📸 Architecture diagram

---

💻 Setup & Usage

Clone the repo:

git clone https://github.com/stoic-dad/project-erasmus.git
cd project-erasmus

Build and run:

docker-compose up --build

Then visit the dashboard at: 👉 http://localhost:8501

---

⚙ Requirements

• Docker (with Compose)
• Optional: Ollama installed locally to run LLMs
• SBOMs and vendor_geo.csv in the proper folder structure (see repo)

---

🧠 Philosophy

This project is a hands-on exploration of combining GenAI, software system complexity, and reasoning. Named after Erasmus — the AI thinker from Dune — it’s built for curious engineers who want to:

• Visualize what software they're running
• Ask natural-language questions about supply chain risk
• Work locally without sending data to the cloud

Let’s make software bill of materials reason about themselves.

---