AI-powered movie recommendations for groups — everyone's taste, one perfect pick.
Picking a movie with a group is painful. Everyone has different moods, different tastes, and no one wants to scroll through a streaming platform for 45 minutes only to give up and rewatch something familiar. Existing recommendation engines optimise for a single user profile — not a room full of people with conflicting preferences.
PopChoice collects each person's preferences through a quick questionnaire, then uses semantic vector search to find a movie that genuinely fits the group's collective vibe — not just a keyword match, but a deep semantic understanding of tone, genre, and taste.
| Setup | Questionnaire | Last Person | Result |
|---|---|---|---|
 |
 |
 |
 |
Each person answers 4 questions about their mood and preferences
↓
All responses are combined into one semantic query string
↓
OpenAI text-embedding-ada-002 encodes the query as a 1536-dim vector
↓
Supabase pgvector finds the closest movie chunks via cosine similarity
↓
OMDB API fetches the poster for the matched movie
↓
Results are presented one at a time
Movies in data/movies.txt are pre-processed at seed time:
- Each movie's text is split into 150-character chunks with a 20-character overlap using LangChain's
RecursiveCharacterTextSplitter— preserving context across chunk boundaries - Each chunk is independently embedded via
text-embedding-ada-002(1536 dimensions) - Chunks and their embeddings are stored in a Supabase
documentstable powered bypgvector
At query time, the combined group preferences are embedded and matched against stored chunks using cosine similarity through Supabase's match_documents RPC function.
| Layer | Technology |
|---|---|
| Frontend | React 18, Tailwind CSS, Vite |
| AI Embeddings | OpenAI text-embedding-ada-002 |
| Vector Database | Supabase + pgvector |
| Text Splitting | LangChain RecursiveCharacterTextSplitter |
| Movie Posters | OMDB API |
popchoice/
├── data/
│ └── movies.txt # Raw movie dataset
├── docs/
│ └── screenshots/ # README screenshots
├── scripts/
│ └── seed.js # One-time DB seeding script (Node.js)
├── src/
│ ├── components/
│ │ ├── SetupScreen.jsx # Number of people + available time
│ │ ├── QuestionnaireScreen.jsx # Per-person preference questions
│ │ └── ResultScreen.jsx # Movie result with poster
│ ├── lib/
│ │ ├── config.js # OpenAI + Supabase clients
│ │ └── recommendations.js # Embedding → vector search → poster fetch
│ ├── App.jsx # App state machine
│ ├── index.css # Tailwind entry
│ └── main.jsx # React entry point
├── supabase/
│ ├── create_db.sql # Table schema (pgvector)
│ └── match_documents.sql # Cosine similarity RPC function
├── .env.example
├── index.html
├── package.json
├── tailwind.config.js
└── vite.config.js
git clone https://github.com/your-username/popchoice.git
cd popchoice
npm installcp .env.example .envFill in .env:
VITE_OPENAI_API_KEY=sk-...
VITE_SUPABASE_URL=https://your-project.supabase.co
VITE_SUPABASE_API_KEY=your-supabase-anon-key
VITE_OMDB_API_KEY=your-omdb-keyRun the following in the Supabase SQL Editor, in order:
-- Step 1: enable the pgvector extension
create extension if not exists vector;Then run:
supabase/create_db.sql— creates thedocumentstablesupabase/match_documents.sql— creates the RPC search function
npm run seedReads data/movies.txt, splits each entry into chunks, generates embeddings via OpenAI, and inserts them into Supabase. Re-running is safe — existing entries are skipped.
npm run devOpen http://localhost:5173.
Append entries to data/movies.txt using the existing format:
Movie Title: YEAR | RATING | RUNTIME | IMDB score
One or two sentence description of the film.
Then re-run npm run seed.
MIT