Hey Product Hunt 👋 My name is Rashad. I'm a software engineer and an M.Sc. Computer Science student at ELTE Budapest. Simurgh AI is my Master's thesis, and this launch is an opportunity to share what I've built so far. The research question I started with: "To what extent can AI agents function as effective stakeholders in architectural decision-making without replacing human engineers?" That last part matters. The goal was never to remove the human from the process. It was to give the human a better room to walk into. The problem I kept seeing in practice Solution architects don't usually fail because their design was wrong. They fail because the CFO was resistant, the legacy team felt threatened, or leadership was surprised in the meeting. The technical work was fine. The political work wasn't done. Most AI tools give you one confident answer. I wanted to explore what happens when you force the AI to disagree with itself first and then let a human make the final call. So I built a Council of Three AI personas: • Legacy Keeper: stability, proven patterns, backward compatibility • Innovator: modern stacks, cloud-native, velocity • Mediator: balanced trade-offs, detects consensus (≥0.8 confidence) They debate your architectural challenge for up to 6 turns using Claude's tool-use API. When they converge, each persona independently writes a complete PRD architecture includes diagrams (Mermaid), tech stack, risks, timeline, and a confidence score. The human evaluates all three. The human approves one. The AI never decides. The stakeholder side The other half of the platform maps the people in the room, not just the architecture. Every stakeholder gets a position on a Mendelow Power/Interest Matrix, a sentiment score from Champion to Blocker, and an AI-generated engagement plan grounded in your actual uploaded project documents via RAG. Because knowing who needs to be convinced, and how, is half the work. The research is still running I built a controlled A/B evaluation framework directly into the platform. Participants complete the same architectural scenario under two conditions: a single-agent baseline vs. the multi-agent Council. After each, they fill out a 7-item trust questionnaire covering risk awareness, technical soundness, balance, and actionability. The central question I'm measuring: does structured multi-agent debate actually improve trust and decision quality, or does it just feel more thorough? I'll share the results here when the study is complete. A few things I'm particularly proud of technically: → Local RAG pipeline: pgvector + FastEmbed, no OpenAI dependency → Async document indexing via Celery (uploads never block the UI) → Fernet encryption for sensitive stakeholder data at rest → 149 passing tests across unit, API, integration, and e2e tiers → Jira + Confluence export via Atlassian OAuth 2.0 → Fully self-hosted, Docker Compose, Apache 2.0 If you're a solution architect, an enterprise architect, or anyone who's ever had a technically sound proposal die in a steering committee, I'd genuinely love to hear how you currently handle that problem. It would help my research, and it would help make this tool more useful. Thank you for checking it out 🙏

Simurgh AI

Three AI architects debate every decision you make.

Three AI architects debate every decision you make.