Building Useful Apps With Small, Local LLMs
summary
Builds a LangGraph.js workflow that uses LangGraph.js with Llama 3 (8B) run via AMA and embeddings from Transformers.js to implement an adaptive RAG pipeline that filters retrieved chunks, checks hallucinations, and checkpoints for human authorization. Smaller local models cut latency (example run 3s vs 15s) and the graph routes technical queries to a vector DB and irrelevant queries to web search, returns accurate answers in demos, and exposes LSmith traces for node-level debugging and rewinding.
AI-written summary · about
chapters (9)
- 0:00Introduction and speaker intro
- 1:27Why small LLMs matter
- 4:39LLM landscape overview
- 8:17Building with future-proof architecture
- 11:27Autonomy spectrum in agentic workflows
- 17:44Retrieval augmented generation basics
- 21:39Building a practical RAG graph
- 24:29Live demo of the RAG system
- 26:34Closing and resources
related talks
40mSurviving the Agentic Hype with Small Language Models
25mBuilding Your own AI Assistant with JavaScript, LLMs, and Rag
48mCrafting your own RAG system: Leveraging 30+ LLMs for enhanced performance
16mReact + WebLLM: Building Intelligent Applications with In-Browser Language Models
01h 19mBuilding an LLM on your Laptop to...
29mUsing LangChain and LlamaIndex to customize LLMs...
27mAI Demystified Unveiling the Intelligence Behind the Machine
31mBuild an Agentic RAG system using Langchain, Ollama and Milvus