Dear Podcast, Can You Help Me? Leveraging OpenSearch and LangChain to ask a podcast for advice
summary
Builds a Retrieval-Augmented Generation pipeline that turns podcast episodes into searchable knowledge: use FFmpeg to extract audio, Whisper → transcribe to get text, LangChain text splitters at ~300-char chunks with 20-char overlap, Hugging Face all-mpnet-base-v2 embeddings (768 dims), store vectors in OpenSearch (KNN enabled) via Aiven, and query with Llama 3 through LangChain for similarity search. The system returns relevant chunks and a synthesized answer, and the demo shows it working end-to-end.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:11Personal journey
- 4:45Workshop and tools
- 6:09RAG patterns explained
- 8:39OpenSearch for this use case
- 10:06LangChain benefits
- 11:09Video to text pipeline
- 12:53Text chunking for LLM
- 13:57Vector embeddings creation
- 15:19KNN and similarity search
- 16:23Querying and retrieval
- 17:04Future possibilities
related talks
35mThe Podcast Pipeline Nobody Sees: Behind the Scenes with AI Agents and AI gen
21mBuilding RAG With OpenSearch ML Plugins: From PDFs To Voice-Enabled Search
21mPodcast to Podcast: Content Localization Using OpenAI API Stack
01h 04mAutomating Podcast Localization via OpenAI APIs
01h 00mLLMs gone wild
01h 02mGenerative AI for the .NET Developer
26mRevolutionizing Search: Unveiling Conversational Search
01h 12mBuild Your Own AI Chatbot with OpenAI & LangChain