Tired of wrestling with 60-page IC datasheets every time you need to write a device driver? In this tutorial, Engineer Ari Mahpour walks you through building a locally hosted LLM-powered datasheet extractor that automatically parses PDF datasheets and outputs structured, LLM-ready data — including text blocks, tables, and figures — so you can hand it off to a flagship model like ChatGPT, Claude, or Gemini to write your device driver libraries for you.
This project uses Python, Docling, Ollama, and the lightweight Moondream vision model to extract and classify everything from register maps to circuit diagrams — all running locally on your machine. Whether you're building for one IC or a hundred, this open-source tool is designed to scale with your embedded workflow, enable AI-assisted pull request reviews, and serve as a queryable database for your entire engineering team. The GitHub repo is live, link below!
⚙️ Resources from this video:
- Ari's Github: github.com/amahpour/datasheet-extractor
- Companion Article (with a bunch of other links): https://octopart.com/pulse/p/building-local-llm-datasheet-extractor-ic-driver-development
✨ Don't Miss Out: Subscribe to our channel for more electronics, supply chain, and industry content: https://www.youtube.com/@Octopart?sub_confirmation=1
⚙️ Browse Octopart for Parts & Tech News: https://octopart.com/
⚙️ TikTok: https://www.tiktok.com/@octopart_official
⚙️ Instagram: https://www.instagram.com/octopart/
#EmbeddedSystems #LocalLLM #DeviceDrivers
0:00 Intro
2:28 Data Extraction Challenges
5:24 Project Overview & Setup ️
8:54 Interactive Notebook Examples
11:22 LLM Analysis Folder Overview
14:05 LLM Data Extraction Challenges
18:26 Table Data Extraction Method
21:28 Application Overview & Library
23:59 Vanilla Docling Extraction
26:17 Schema Definition & Metadata
29:41 GPU Utilization & Olama
32:51 Local LLM Image Classification ️
36:10 Pipeline Application Process ️
39:55 Comparing Model Performance ️
42:20 Multi-Tiered LLM Approach 🪜