Chaining, Smithing, and Graphing LLMs on Google Cloud
summary
A pipeline that turns a 30,000‑character blog post and photos into ready captions using LangGraph to orchestrate steps, Gemma 2 (9B) for caption generation and Gemini 1.5 Pro for large‑context summarization, with LangChain for prompt/parse helpers and LangSmith for observability. It summarizes 30,000 chars → 1,700 then generates ~1,000‑char image descriptions, offers five caption proposals in ~8–10s per step, supports a human‑in‑the‑loop refine cycle, writes the chosen caption to file and integrates it into the blog successfully.
AI-written summary · about
chapters (14)
- 0:00Introduction and speaker intro
- 1:28Skepticism to Gen AI journey
- 2:54Okinawa coral project inspiration
- 5:26Talk overview and structure
- 6:55Website structure and setup
- 8:21Demo of automated caption generation
- 12:55LangGraph workflow explanation
- 15:03Code walkthrough and workflow details
- 16:50LangGraph state and memory saver
- 17:10Gemini and Gemma models overview
- 18:13LangChain for pipeline construction
- 19:16LangSmith for debugging and observability
- 20:03LangSmith demo and metrics
- 21:06Q&A session
related talks
25mFrom LangChain to LangSmith and LangGraph: the Future of AI
01h 50mChatGPT & LLMs: A Practical Starter's Guide to Language Chains
28mThe Present and Future of AI-Enabled Knowledge Graph Generation
51mGemini, Google's Large Language Model
29mUsing LangChain and LlamaIndex to customize LLMs...
27mBuilding Useful Apps With Small, Local LLMs
51mGemini, Google's Large Language Model
01h 34mA Practical Starter's Guide to building LLM based projects