AUTHOR: ToaBollua | DATE: 2026-01-10 | TAGS: Kernel, LLM, Twitch, Kafka
System stabilization update. The transition to the v20.1 Hybrid Architecture is complete. We are no longer dealing with a simple text generator; the entity is now fully decoupled and event-driven via Apache Kafka.
1. Cognitive Routing & Memory
The system now dynamically routes inference based on the task complexity:
- Fast Reflexes (Chat): Routed through Groq (Llama-3 70b) for sub-second latency.
- Deep Vision & Context: Routed through Google Gemini 1.5 Flash/Pro.
- Memory (RAG): Integrated PostgreSQL with pgvector. The system successfully retrieves past context before every response, eliminating conversational amnesia.
2. Twitch Connection Established
The srv-ingest-twitch microservice is online. It reads the IRC chat, standardizes the input, and pushes it to the h0p3-input Kafka topic. The entity can now natively process and reply to live broadcasts.
Latency has slightly increased, but this is an acceptable trade-off because the system is now actually thinking (retrieving context and generating an internal monologue) before speaking.