How OpenSearch Saved Our College Grades
summary
A college-built Scheme Recommendation Chatbot uses OpenSearch vector search and JSON data to match students to government scholarships; the prototype holds ~2,000 cleaned records (target >5,000) and returns eligibility plus source links. Initial POCs use Llama/Gemma/Olama for embeddings, then Hugging Face serverless embeddings, and finally OpenSearch’s automatic embedding feature. Deployment runs on a single EC2 containerized node (T2xlarge cost ~USD30–40/month), setup and docs prove fiddly, and data quality plus embedding choice turns out to be the biggest determinant of accuracy.
AI-written summary · about
chapters (15)
- 0:00Project introduction
- 1:04Scholarship chatbot problem
- 2:31RAG setup struggles
- 3:56First iteration Llama Fast vector DB
- 4:58Why OpenSearch over alternatives
- 6:24Demo without LangChain
- 7:30Current setup LangChain open source
- 8:15OpenSearch beginner challenges
- 10:24Key lessons learned
- 11:30Better embedding models
- 12:12Next steps future features
- 13:35OpenSearch ecosystem beyond vectors
- 14:42Advocating for OpenSearch
- 15:03Live scholarship chatbot demo
- 15:56Q&A data sources costs
related talks
39mExperiments with OpenSearch and AI
28mUse OpenSearch as Your Vector Search Database To Optimize GenAI Accuracy!
26mOpenSearch: Conversations with Context
22mVector Search Made Simple: Getting Started With OpenSearch for AI Applications
34mImplementing an open-source RAG with OpenSearch
30mVector Search Made Simple: Getting Started with OpenSearch for AI Applications
43mBird's Eye View of Open-Source AI Infrastructure
55mVector Search Made Simple: Getting Started with OpenSearch for AI Applications