Computers That “Get It”: Using LLMs to Give Software Semantic Understanding
summary
Claims that LLMs plus vectors let software actually map meaning, demonstrated by a production deployment for American Cancer Society: a mobile app adding semantic search and content recommendations alongside existing keyword results (hybrid keyword+semantic). Implementation uses sentence embeddings (all-miniLM-L6-v2 → 384-dimensional vectors, model <0.1 GB) stored in Chroma linked to a relational CMS; semantic matchmaking for patient-volunteer pairing runs live and avoids RAG generation in high-liability contexts to eliminate hallucination risk.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:33Why computers struggled
- 6:55Vector databases explained
- 12:20Real-world semantic search
- 18:25Semantic matchmaking demo
- 22:20Choosing the right tools
- 25:54Key takeaways
- 26:44Q&A session
related talks
49mLLM—Search Engine Killer?
46mLLM—Search Engine Killer?
24mConversational Search with LLMs
28mEnhancing LLMs with Graph Technology
15mFrom Words to Data: Using LLMs to Turn Questions into SQL
22mPractical Strategies for Implementing Domain-Specific LLMs
07mLightning Talk: Grounding LLM to Avoid Hallucinations Using Vector S...
44mGive Your LLMs a Left Brain