Overview

A stateless AI reasoning engine that processes questions against PDF documents using Google Gemini 1.5 Flash. Leverages the massive 1-million-token context window for the “Combined Mind” approach, enforcing a strict retrieval-only policy with zero external knowledge.

Features

• Strict Mode: Zero-knowledge policy - answers ONLY from provided PDF context
• Multimodal Input: Supports both text questions and image-based questions
• Source Citations: Every answer includes file name and page number
• Question Type Detection: Specialized reasoning for Multiple Choice, True/False, Identification, and Long Answer
• Structured Responses: Clean JSON output with reasoning steps and confidence scores
• High Performance: Optimized for low latency with transient logging

Key Components

File	Purpose
app/api/brain/route.ts	Main API endpoint handler
lib/brain-types.ts	TypeScript type definitions
lib/brain-prompts.ts	System prompts & context formatting
lib/brain-client.ts	Frontend client utilities