Working with Memory and Dialogue History in LLM 🧠

Another tech support chatbot is being rolled out, the user is ranting about their problem, and on the third message the neural network asks: "What is your name and how can I help you?".

The reason is trivial: LLMs are stateless by nature. The model's "memory" is just an architectural crutch, and the problem is usually solved "head-on": they take LangChain, collect all previous messages and throw them into the prompt.

🗓 April 14 at 18:00 Moscow time at "Assembly Point" we will analyze the memory architecture for LLM assistants and methods for adequate context management.

Required background for understanding the material:
🔵Basic Python syntax (classes, dictionaries, functions).
🔵Understanding HTTP API and basic LLM concepts (what a prompt and token are).
🔵 Superficial familiarity with LangChain abstractions.

Access via bot: t.me/TScompiler_bot