Building AI-Powered Apps with NodeJS and Gemini | Armen Vardanyan #JSConfAM25
summary
A Node.js demo that indexes product JSON using Google Gemini embeddings (≈3076‑dim), stores those vectors, and performs cosine‑similarity semantic search for multilingual queries via Google SDK calls and a lightweight cosine library. It uses structured JSON output from the LLM to generate UI filters/actions and trains a small logistic‑regression classifier on embeddings to predict valid queries, demonstrating relevant multilingual results at low cost (~$0.15 per million tokens).
AI-written summary · about
chapters (9)
- 0:00AI Node.js apps
- 2:13AI myths in JS
- 6:32Embeddings and similarity
- 10:31Semantic search demo
- 15:52Structured LLM outputs
- 19:11Embeddings and regression
- 26:23Chatbot optimization
- 27:48Node.js AI features
- 29:12Q&A
related talks
37mIntroduction to the Gemini API for Web Apps
01h 01mBuilding Multimodal Apps with Gemini 2.0 and Gemini SDK
01h 45mAI Engineering with the Google Gemini 2.5 Model Family
36mTypescript, You and AI Search
18mIntroduction to Gemini APIs and AI Studio
56mBuilding Multimodal Apps with Gemini 2 0 and Gemini SDK
01h 48mBuilding Conversational Agents
51mGemini, Google's Large Language Model