// the find
NeuralNine/neuralintents
A simple interface for working with intents and chatbots.
neuralintents wraps a Keras bag-of-words intent classifier in a three-line API: define tags/patterns/responses in a JSON file, call fit_model, and the BasicAssistant class trains an intent model and maps matched intents to either canned responses or real Python functions. It's aimed at people who want a toy chatbot or voice-assistant skill running fast, not at anyone building production NLU.
The method_mappings pattern is the one genuinely useful idea here — binding an intent tag directly to a Python function means you can turn 'what are my stocks' into an actual function call instead of a scripted reply, which is the difference between a chatbot demo and something you can wire into real code. The API surface is tiny (fit_model, save_model, process_input) so you can go from zero to a working classifier in under ten lines. The examples directory covers four distinct scenarios including overriding the underlying network architecture, so there's at least a path past the default model if you need one.
The README's own first line is 'still in a buggy alpha state,' and the last commit was in December 2024 — that's the maintainer telling you not to trust it, with no indication it's being actively stabilized. This is 2016-era bag-of-words intent classification with a from-scratch training step (fit_model(epochs=50)) every time you touch the intents file; there's no incremental training, no confidence threshold handling visible in the API, and no actual NLU — it's keyword/pattern matching dressed up as a neural net. There's no requirements.txt or pinned Keras/TensorFlow version anywhere in the tree, so a routine TF upgrade can silently break fit_model or change saved-model compatibility. No tests exist in the repo, so there's nothing catching regressions between releases, which matters more given the alpha warning.