interviewed at the Artificial Intelligence Conference, London 2018 – Amy Heineike
summary
Company builds NLP that drafts reports as well as a junior analyst, combining a hybrid template‑extractive‑abstractive pipeline with pointer models that can fall back to extraction when generation fails. Systems flag significance by measuring edit diffs in 10‑K filings and surge signals across sources, surface entities via lightweight knowledge graphs, and score candidate sentences for coherence. Evaluation relies on human judgments beyond ROUGE. Engineering uses Elasticsearch, CI/CD and optimization-heavy ranking for production scale over thousands to tens of thousands of documents.
AI-written summary · about
chapters (5)
related talks
43mKeynote: Applied NLP in the age of Generative AI
27mBuilding Generative AI features
35mThe Podcast Pipeline Nobody Sees: Behind the Scenes with AI Agents and AI gen
36mBuilding Agentic AI Assistants: Lessons from productionising agents in Preclinical R&D
01h 24mStay Ahead in AI with Product Engineering and AI Native Tools
24mDocumentation and AI: How to write right now
07mInterview
28mAI’s Next Great Challenge: How to create AI that truly understands us?