The final camera-ready ACM version of my IAIT2026 paper #6099 is now complete:
DLBP: Deterministic Blackboard Pipelines with Specialized LLM Knowledge Sources — A Generalizable Architecture for Intelligent Multi-Stage Reasoning
This paper presents the Deterministic LLM Blackboard Pipeline, or DLBP, as an architectural pattern for building controlled, staged, auditable LLM reasoning systems.
The core idea is simple:
Instead of treating an LLM application as one large prompt chain, DLBP separates reasoning into deterministic pipeline stages, each supported by specialized LLM knowledge sources and structured blackboard state. This makes the reasoning process easier to inspect, govern, reproduce, and improve.
The paper argues that many useful LLM systems do not need more magical prompting. They need better architecture.
This work grew out of my ongoing experiments with Ruby, Rails, blackboard-style systems, and multi-stage LLM reasoning pipelines.
The final ACM camera-ready version is attached here for anyone interested.
IAIT2026_ACM_Paper_6099_Final.pdf (606.5 KB)