Veritas AI

Overview

A production-grade AI chat platform built on Cloudflare infrastructure. Combines Claude-powered reasoning, web search capabilities, and visual architecture generation into a single platform. Implements deterministic workflows with automatic error recovery.

Features

• Multi-step AI workflows with deterministic retries and auto-repair loops
• Persistent chat memory with full session state in D1
• Real-time response streaming to clients
• Mermaid-based architecture diagram generation with validation
• Parallel multi-engine web search with automatic synthesis
• Production monitoring with measurable performance metrics
• Intent-aware routing between four execution modes
• Zero-warmup horizontal scalability across edge locations

Architecture

The system follows a clean request-response pipeline optimized for edge computing:

Request Flow: Frontend → Cloudflare Workers → Workflows → Workers AI → D1 Database → Streaming Response

Intent Classification: The initial Worker analyzes each query and routes to one of four execution modes:

• Normal: Conversational AI for general questions
• Search: Multi-step web search synthesis
• Reasoning: Chain-of-thought decomposition for complex problems
• Developer: Architecture diagram generation with Mermaid + verification loop

Data Persistence: All chat history and workflow state persists in Cloudflare D1 (SQLite), enabling multi-turn conversations with full context.

Error Handling: The system includes an auto-repair loop for malformed outputs. If Workers AI generates invalid JSON, the workflow automatically attempts correction rather than failing.

Tech Stack

• Cloudflare Workers
• Cloudflare Workflows
• Workers AI
• Cloudflare D1
• Cloudflare R2
• Next.js 14
• React
• TypeScript
• pnpm
• Node.js LTS