ABCD Data Dictionary Chatbot
A Shiny app that lets researchers search the ABCD data dictionary with natural-language semantic queries.
Finding the right variable in the ABCD dataset can mean scanning thousands of cryptically named fields by hand. This tool makes that search conversational: type “screen time on weekends” or “BMI” and it returns the dictionary entries that are semantically closest, ranked by similarity — so the measures you need surface even when you don’t know their codes.
It’s a compact R + Python app: a Shiny front end over a MiniLM sentence-embedding model quantized to run on CPU, with precomputed embeddings and Parquet-backed dictionaries for fast lookups. It tracks multiple ABCD dictionary releases and detects the version automatically.