// the find
llSourcell/tensorflow_chatbot
Tensorflow chatbot demo by @Sirajology on Youtube
A 2016-era wrapper around a Cornell Movie Dialogues seq2seq chatbot, built for a Siraj Raval YouTube tutorial, with a small Flask UI bolted on for demoing the trained bot in a browser. Useful only as a historical/educational reference for pre-attention seq2seq chatbots, not as a starting point for anything you'd actually ship.
It's a complete pipeline — data preprocessing, training, inference, and a working Flask+jQuery chat UI — so you can see every stage from raw dialogue corpus to a chatting webpage in one repo. The .ini-based config (seq2seq.ini) cleanly separates train/test mode and hyperparameters from code, which is a reasonable pattern for a tutorial project.
Built against TensorFlow r0.12's tf.contrib.legacy_seq2seq, which was deleted outright in TF 2.x — this will not run on any TensorFlow installed in the last 6+ years without pinning an ancient, unsupported Python/TF combo. No tests, no CI, and the author's own README says most of the actual modeling code is copied from someone else's repo (suriyadeepan) with a thin wrapper on top. The bundled vocab files and working_dir artifacts are checked into git, which is a red flag for repo hygiene, and there's zero attention/transformer architecture — this predates that shift entirely, so the conversational quality will be noticeably worse than anything from the last several years.