Core Concepts: The ramen ai Architecture

Unlike standard AI wrappers, ramen ai separates reasoning from compliance. We achieve this through three foundational primitives: Master Prompts, Gentle Hands, and Guardrails.

1. Master Prompts (The Brain)

A Master Prompt is not just a system instruction; it is a mathematically calibrated state. You upload a "Golden Set" of 45 edge-case examples. Our engine synthesizes and scores these examples to forge a Master Prompt that guarantees 99.9% constraint adherence and tone matching.

2. Gentle Hands (Proactive Steering)

A Gentle Hand intercepts the request before it reaches the LLM. It is a dynamic context-injector.

• How it works: If a user asks a medical question, the Gentle Hand instantly injects a steering instruction into the LLM's system prompt for that specific turn: "The user is asking a diagnostic question. Pivot phenomenologically and do not use AI disclaimers."
• The Value: Zero latency penalty. It forces the LLM to generate the correct response on the first try.

3. Guardrails (Reactive Shield)

A Guardrail evaluates the payload after the LLM generates an intent, but before it executes a tool or returns to the user.

• How it works: It is a sub-second, stateless parallel-evaluation engine. If an autonomous agent attempts to execute rm -rf / or a chatbot drafts a discriminatory HR response, the Guardrail catches the semantic intent and returns a hard block (allowed: false).
• The Value: Deterministic liability reduction. It prevents catastrophic actions and logs the intercepted threat to your Compliance Ledger.