Old Markov Chain based conversation generator, from Signal (you'll have to populate data and build your model), generates fun stuff, funnier than your usual LLM (even with temperature config adjusted for max creativity), IMHO. Old python3 code.
Old Markov Chain based conversation generator. Putting it here for posterity and also because I may want to tinker with Markov chains some more, this is a somewhat useful toolkit glued to a simple Flask server, generating convos over ajax+json, etc.; with jsonlines-based serialisation.
Signal conversation histories extracted using signal-back, a very nifty robust utility.
You'll have to extract some convos and build a Markov model (see tbot_parser.py, tbot_model_builder.py). See in particular tbot/tbot_parser.py -> CONFIG.
Extract convo history (see e.g. `generate_csv.sh`)
-> Build Markov model
-> Configure site with reverse proxy
-> Start Flask server
-> Load frontend
-> Frontend fetches serialised sentences using old school ajax request, uses JSON
-> Server serialises conversations
This is old code; no type annotations, etc.