🐍 Miru

👁️ Multimodal reasoning tracer — see what models see, and why they decide.

---

👁️ Meaning

Miru (見る) — to see, to observe.

Not just outputs — but perception and reasoning.

---

🚀 What it is

Miru is a multimodal explainability engine:

• Input: image or document + question

• Output:

  • answer
  • reasoning trace
  • attention visualization

---

❗ The problem

Multimodal models are black boxes:

• No visibility into reasoning
• No auditability
• No explainability

Critical issue for:

• compliance
• medical
• enterprise AI

---

🧠 What you learn

• Vision-language models (VLMs)
• Cross-attention mechanisms
• Saliency & interpretability
• Multimodal reasoning

---

⚙️ Stack

• 🐍 Python (FastAPI backend)
• 🎨 Visualization layer (attention maps, overlays)

---

🚀 Quick Start

uvicorn miru.main:app --reload

---

🎯 Vision

Make AI reasoning visible.