Consensia

Multi-agent reasoning and evaluation for software engineering.
Because truth deserves more than one mind.

Prototype & Approach

Our prototype allows users to create multiple AI personas with different software-engineering and business roles — such as CTO, Software Architect, Senior Developer, QA, SRE, Security Engineer, as well as Product Manager, Finance/CFO, and Operations/Management. Each agent answers a technical or strategic question; a Judge LLM analyzes all outputs and scores them for consistency, fairness, and reasoning clarity. The judge’s consensus is then validated against human-tagged data.

System Concept

Multi-agent discussion leading to a final, explainable verdict generated by the Judge LLM, combining text reasoning with evidence from tools.

Stakeholders

Technical roles (CTO, Architect, Dev, QA, SRE, Security) balanced with business roles (Product, Finance/CFO, Ops/Management) to reflect real enterprise trade-offs.

Architecture

Event-driven orchestration for scalability and real-time updates. Frontend in React, backend with Python, automation via n8n.

Team & Advisors

Project Team

  • Amir Hossein Ahani
  • Ahmed Hatem Haikal
  • Türker Köken
  • İrfan Hakan Karakoç
  • Mehmet Hakan Yavuz

Supervisors & Advisors

Supervisors

  • Mert Bıçakçı
  • İlker Burak Kurt

Advisor

  • Prof. Anıl Koyuncu
    Department of Computer Engineering
    Bilkent University

Innovation Expert

  • Haluk Altunel

Project Roadmap

Research → UI design → Features → LLM self-judging automation → Prototype → Analysis & statistics → Evaluation → MVP → Publication