Project Walkthrough Video
This was my final research assignment for the Artificial Intelligence course at Kristianstad University (HKR). Big thanks to my professors Charlotte Sennersten and Craig Lindley for their guidance throughout the course. 🙏
Project Overview
As Industry 5.0 moves toward Industry 6.0, human centered AI is increasingly designed to augment rather than replace human decision makers. Healthcare is one of the most critical domains for this collaboration. This project designs a hybrid multi agent architecture for a Healthcare Clinical Decision Support System (CDSS) that integrates AI reasoning with clinical expertise.
Goal
Transform complex patient data into actionable, explainable recommendations that improve diagnostic accuracy and treatment appropriateness, while keeping the clinician in full control. The system augments clinical judgement, reduces diagnostic errors, and mitigates cognitive overload.
The Multi Agent Architecture
The CDSS is composed of eight specialized agents coordinated by a Meta Agent, with the Human Clinician retaining final authority:
- Data Management Agent · integrates EHR records, labs, imaging, and observations.
- Risk Assessment Agent · estimates disease probabilities and outcomes.
- Knowledge & Guideline Agent · applies clinical standards and ontologies.
- Diagnostic Reasoning Agent · infers likely diagnoses.
- Treatment Planning Agent · evaluates interventions and prognosis.
- Ethics & Safety Agent · checks compliance, fairness, and risk constraints.
- Explanation & Interaction Agent · communicates rationales and uncertainty.
- Human Clinician Agent · provides oversight and final decision authority.
- Meta Agent · coordinates outputs and resolves conflicts.
Hybrid Inference
To illustrate the system in action, the project focused on a clinical scenario: detection and treatment of suspected bacterial infection with potential sepsis progression. The reasoning cycle combines five paradigms:
- Deduction for clinical compliance
- Induction for probabilistic learning
- Abduction for diagnostic explanation
- Utility reasoning for treatment optimization
- Meta reasoning for conflict arbitration