From Natural Language to Structured Solr Queries using LLMs
summary
Prototype converts natural-language questions into structured Solr queries by using an instruction-tuned LLM to do two steps: JSON field-extraction from a Solr schema/dictionary and query expansion (synonyms/variations) then assemble filters into a Solr query. Implementation uses the dpy library for prompt modules and Apache Solr for index/testing with an OECD statistical dataset; experiments show useful disambiguation but frequent hallucinated fields/values, token-limit latency, and no production benchmarks yet.
AI-written summary · about
chapters (10)
- 0:00Introduction and context
- 2:08Use case and LLM basics
- 4:00Lexical search limitations
- 7:13LLM-enhanced search strategy
- 10:04Architecture and workflow
- 16:06DSPy framework exploration
- 18:31Findings and benefits
- 23:07Challenges and limitations
- 31:31Path to production
- 34:19Q&A and conclusion
related talks
24mConversational Search with LLMs
42mText-to-Struct: Fine-tuning SLMs for Query Intent
45mEnd-to-End Semantic Search with Apache Solr 9.8 LLM Module
15mFrom Words to Data: Using LLMs to Turn Questions into SQL
51mQuery Understanding With Large Language Models: Techniques and...
01h 30mUsing LLMs to improve your Search Engine
46mBeyond Basic Prompting: Supercharging Open Source LLMs with LMQL's Structured Generation
49mLLM—Search Engine Killer?