Building a Recommendation System with a Hybrid Search From Scratch
Talkat PythonAsia 2026 · on Saturday, 2026-03-21· at 13:15· Yuchengco Hall 5th Flr. Y507 (Workshop Room 1)·
summary
Builds a working recommendation demo that combines keyword BM25 scoring and vector similarity into a hybrid ranker, ingests sample movie data into MongoDB Atlas + Voyard embeddings, creates a vector index (dim ~1000) and a full‑text index, runs on a Django backend + LangChain pipeline, and returns merged scores with optional reranking; notes practical limits (100 free Atlas projects, 3‑node free clusters, 200M tokens on Voyard) and shows the demo working after minor deployment hiccups.
AI-written summary · about
chapters (9)
- 0:00Introduction and setup
- 5:46MongoDB overview and setup
- 25:57Recommendation systems basics
- 35:39Hybrid search concept
- 40:15Demo environment setup
- 46:30Chunking methodology
- 59:51Data ingestion and indexing
- 1:01:39Vector search implementation
- 1:06:01Q&A and conclusion
related talks
57mHands-On with LLM-Powered Recommenders: Hybrid Architectures
01h 31mHands on - Build a Recommender system
50mMachine Learning + Graph Databases for Better Recommendations
55mBuilding a Recommendation Engine with Machine Learning Techniques
01h 21mUnlocking Personalization: A Deep Dive into Modern Recommendation Algorithms
23mML-Powered Search & Recommendation 101
37mSearch and Recommendations: 3 Sides of the Same Coin #bbuzz
37mA primer on recommendation systems.