Core Architecture

• Atomic Reasoning: Decomposes problems into verifiable atoms with bottom-up synthesis.
• External Memory: Utilizes sentence-transformer embeddings for long-term conversational recall.
• Intelligent Routing: A zero-shot classifier automatically selects the optimal path (direct, memory, or reasoning).
• Calibrated Confidence: Responses include a confidence score (e.g., Confidence: X%).

Framework Comparison: HANERMA vs. LangChain