// the take
A portfolio/demo project showing how to wire Kafka, TensorFlow.js, MongoDB, and Node.js into a stock prediction pipeline. It streams CSV data through Kafka topics, trains an LSTM-style model on historical prices, and runs predictions in a parallel consumer. This is a learning exercise, not a production system.
The architecture diagram actually reflects the code — producer, two consumers, and MongoDB are all present and wired correctly. Using TensorFlow.js server-side keeps the ML and streaming in the same runtime without a Python sidecar. The separation of a db-writing consumer from a prediction consumer is a sensible Kafka pattern. MinMax normalization before training is at least the right instinct for time-series price data.
Last commit was February 2021 — five years stale, and the dependencies will be a mess of CVEs. Only two CSVs in the dataset folder (A.csv, HPQ.csv), so 'real-time' means replaying static files, not live market data. The model predicts normalized values and prints raw floats with no denormalization or price context, making output useless for any actual decision. No error handling, no tests, no env config — hardcoded values scattered across config.js and utils.json. The author admits the client side was abandoned.
We read the README, the directory tree and a key source file in June 2026. Nobody pays to be listed. How we judge →