Building an enterprise NLSE ...
summary
Built an enterprise QA pipeline that plugs Facebook’s DrQA reader into an ElasticSearch retriever over Office/PDF docs (ingested with Apache Tika), producing a deployable search UI that shows answer spans plus document previews and filters. Initial TF‑IDF retriever hits ~76% P@5 on manuals; switching to ElasticSearch raises reader success to ~84% with an end-to-end F1 ≈ 42%, and 94% of first results contain the right context. Works in practice (operators report ~5 minutes saved) but often returns contextual pages rather than exact spans.
AI-written summary · about
chapters (10)
- 0:00Introduction to the project
- 1:31Enterprise search challenges
- 3:51Key barriers to implementation
- 7:25Natural language and chatbots
- 9:26Search and chatbot integration
- 13:54The PMI technical use case
- 21:08Optimizing search performance
- 24:53Takeaways and user experience
- 26:47Future work and BERT
- 28:30Q&A and closing remarks
related talks
28mFrom Tables to Answers: building QA System for In-Document Searches
34mExposing Dark Data in the enterprise with custom NLP
01h 30mBuilding a Search Engine
01h 28mBuilding a Semantic Search Engine
33mDeep research at scale
33mA Practical Approach To Semantic Search
45mLessons learned in bringing a RAG chatbot with access to 50k+ diverse documents to production
28mFinding answers faster with intelligent search