An LLM Toolchain for Open Data and NGSI-LD
summary
An LLM pipeline converts raw website and PDF text into NGSI-LD entities through a four-step LLM chain that extracts facts, merges properties, creates graph relationships and instantiates missing entities into an NGSI broker. The project yields thousands of entities, 100+ types and powers real apps (e‑bike rentals, audio guides, hiking paths, geotagged comments). Open models (Fi4‑mini) match costly GPT on many metrics when tuned locally on 32 GB hardware, but entity disambiguation and hallucination mitigation remain necessary.
AI-written summary · about
chapters (4)
- 0:00Introduction and project overview
- 3:45NGSI-LD standard explained
- 8:02LLM toolchain for data extraction
- 11:59Final considerations and Q&A
related talks
44mGive Your LLMs a Left Brain
52mGive Your LLMs a Left Brain
01h 19mBuilding an LLM on your Laptop to...
01h 49mAI engineering with open access LLMs that lie, curse, and steal
24mDemocratizing data access through LLM
36mTechniques to avoid common LLM pitfalls
33mDeep research at scale
15mOpenAssistant's Opensource LLMs Saga