ACRE — Adaptive Cognitive RAG Engine
Most RAG chatbots retrieve a few chunks and paste them into an answer. ACRE goes further, using four coordinated modules to reason like a researcher would. The Query Planner decomposes complex questions into atomic sub-questions. The Graph Chunker builds a semantic knowledge graph from ingested documents instead of flat text chunks, using LLM-based concept extraction. The Retrieval Critic scores every retrieved chunk on relevance, coherence, freshness, and specificity before it reaches the model. The Conflict Resolver detects contradicting information across sources and explicitly resolves it, citing which source won and why. Every answer is also rendered as a clean, topic-agnostic flowchart generated directly from the model's own reasoning — turning any document into a visual map for fast revision, not just another chatbot transcript. Built on Fireworks AI (Qwen3.7 Plus) and fully containerized with Docker.
Category tags:
.png&w=96&q=75){kind=small}
For Innovators & Creators