Querying Databases with Function Calling
summary
Claims a function-calling API layer improves natural language database queries, builds a Weaviate query agent and transform agent, and benchmarks Gorilla LLM and others on 315 synthetic queries, reporting 74% exact match for a top model. Results show structured outputs let models emit valid calls but models struggle with many optional arguments, often omit function calls, and fail on search vs text-filter distinctions, so the approach is promising technically but not fully reliable for complex, multi-argument APIs.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:35Merge search engines with databases
- 3:41Function calling in queries
- 5:46Structured outputs and MCP tools
- 7:52Weaviate Gorilla and evaluation
- 10:38Synthetic schemas and data generation
- 12:04Benchmark results and performance
- 13:49Structure generation and future research
- 15:57Next steps in research
- 17:42Agentic vs compound AI systems
- 19:48Foundation models for agentic databases
- 21:10Weaviate’s agents layer
- 22:57Q&A
related talks
14mPutting the "fun" in function calling
23mBuild Natural Language-Powered Apps with Select AI
31mCreating Context: Loading your life into a vector database
01h 00mWorkshop: Getting Data Ready for AI - With Unified GraphQL &SQL Endpoints
47mHow to query data using natural language - intro to AI features in Oracle 23ai
31mVue, You and AI Search
24mHow Generative AI Apps Use APIs
40mLLMs & Data Warehouses: Reliable AI Agents for Business (Keynote)