Next-Gen Documentation with AI
summary
Claims that LLMs can automate API and code docs, but demo shows mixed results: a Streamlit UI pulls Python files, calls open-source models (Mistral, Llama, IBM Granite and OpenAI for baseline), chunks code by tokens (context windows noted at 32k–128k–1M), and outputs generated docs compared to handwritten ones. Evaluation uses automated metrics, an LLM-as-judge and human review via Aria; hallucinations and low scores appear, on‑prem deployment runs on OpenShift AI to preserve data and scale.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:03Documentation evolution
- 6:50Generative AI overview
- 8:57How LLMs work
- 13:32Context window in LLMs
- 17:07LLM use cases
- 20:20LLMs for documentation
- 22:04Implementing an LLM application
- 25:15Third-party LLMs
- 28:04Local LLM deployment
- 30:33On-prem LLM solutions
- 34:33Demo: auto docs
- 42:02Evaluating LLM docs
- 44:31Challenges and limitations of LLMs
- 45:35LLMs future in docs
- 47:24Call to action
related talks
49mNext-Gen Documentation with AI
35mHarnessing AI for API Business Documentation in a Devportal CMS [...]
37mDocumenting your open source projects for machines
18mLLMs and GenAI: A Guide to Building Your Own RAG Model for Document Interaction
43mLarge Language Models for Devs: from zero to your first LLM application
33mOpening Keynote: LLMs, between fantasy and reality - the poin...
47mBuilding and Testing AI-Agent Powered LLM Applications: A Live Demonstration
27mAI Demystified Unveiling the Intelligence Behind the Machine