Network Configuration Parser

Phase 1 — Knowledge Base Ingestion

← Back to Projects

The Insight

Developing…

Quick Facts

What This Is

A framework for parsing and analyzing device configurations across multiple networking vendors. The project bridges the gap between unstructured legacy CLI data and structured intent-based models.

Phase 1 focuses on building the Knowledge Base: ingesting thousands of pages of vendor manuals (PDFs) and converting them into high-fidelity Markdown. This creates a searchable, semantically-indexed reference that AI agents can use to accurately interpret vendor-specific syntax.

Problem It Solves

• Syntax Fragmentation: Every vendor (Cisco, Juniper, Arista, Nokia) has a different CLI structure for the same protocol (OSPF, BGP).
• Manual Translation: Engineers spend hours cross-referencing manuals to convert a “design” into “commands.”
• Data Silos: Configuration state is locked in text files rather than queryable databases.

Features

• Layout-Aware Ingestion: Uses pymupdf4llm and MinerU to preserve tables, hierarchies, and code blocks from vendor PDFs.
• Semantic Search: Vector store integration (ChromaDB) enables “Search by Intent”—find how to configure a feature by describing its function, not just its command.
• Unified Model Target: Designed to output standardized YANG/OpenConfig models, enabling downstream tools like the Network Modeling & Configuration Library to consume legacy state.
• API-First: FastAPI backend provides endpoints for ingestion pipelines and search queries.