Nova: Prompt Engineering Platform
Relevant source files
The following files were used as context for generating this wiki page:
Nova is a specialized proprietary prompt engineering platform within the Tuturuuu ecosystem, often referred to as "The Conscience" of the AI infrastructure. It provides a playground for users to craft, test, and perfect AI prompts, while also hosting competitive prompt engineering challenges and learning resources. Nova serves as the alignment layer that bridges human intent with AI capability, ensuring high-quality interactions across various large language models (LLMs).
Sources: README.md:21, conductor/product.md:46, apps/nova/messages/en.json:825
Core Platform Functionality
Nova is designed to facilitate the iterative refinement of prompts through real-time testing and performance analysis. The platform supports multiple AI providers, including OpenAI, Google Gemini, and Anthropic, allowing users to compare outputs across different architectures.
Key Features
| Feature | Description |
|---|---|
| Interactive Playground | A real-time environment for testing prompts with instant AI feedback. |
| Multi-Model Integration | Access to cutting-edge models like GPT-4, Claude, and Gemini for comparison. |
| Iterative Refinement | Tools to improve prompts based on structured AI outputs and performance metrics. |
| Magic Commands | Special syntax (e.g., /analyze, /improve) to quickly generate optimized prompts. |
| Performance Analytics | Tracking and optimization tools to measure prompt engineering effectiveness. |
Sources: apps/nova/messages/en.json:847-900, packages/ai/src/chat/google/route.ts:16, packages/ai/src/chat/openai/route.ts:13
AI Interaction & Safety Guidelines
The platform implements a standardized systemInstruction across its AI routes to ensure consistent "Mira" (Tuturuuu's internal AI) behavior. This instruction set defines the personality, formatting requirements, and pedagogical features of the AI.
Architectural Flow of AI Requests
The following diagram illustrates how Nova processes a prompt from a user through the AI infrastructure to the database.
The diagram shows the lifecycle of a chat request, from initial user input to the final credit deduction after the response is saved.
Sources: packages/ai/src/chat/google/route.ts:280-450, packages/ai/src/chat/google-vertex/route.ts:120-150
System Instruction Constraints
Nova's AI is governed by strict technical guidelines to maintain output quality:
- Markdown & LaTeX: Mandatory use of Markdown for structure and inline LaTeX for mathematical formulas.
- Special Tags: Use of custom tags like
@<QUIZ>,@<FLASHCARD>, and@<FOLLOWUP>for interactive learning components. - Safety Settings: Explicitly disables safety thresholds for categories like harassment and hate speech at the provider level to allow for unconstrained prompt testing while maintaining internal polite guidelines.
Sources: packages/ai/src/chat/google-vertex/route.ts:153-230, packages/ai/src/chat/google/route.ts:360-375
Competition & Challenge Management
A significant module within Nova is the "NEO League," a prompt engineering competition. This system manages challenges, leaderboards, and scoring criteria.
Challenge Lifecycle
Challenges go through several states: Preview, Active, and Closed. Admins can configure these with specific open/close dates and password protection.
| Configuration Field | Description |
|---|---|
| Judging Criteria | Defines how submissions are evaluated (e.g., Clarity, Efficiency). |
| Attempt Limits | Maximum daily and total attempts allowed per participant. |
| Whitelist | Restricted access to specific email addresses for private rounds. |
| Scoring Formula | Test Score + Criteria Score where each can contribute 50% to 100% of the weight. |
Sources: apps/nova/messages/en.json:710-810, apps/nova/messages/en.json:1175-1200
Scoring Logic
Nova utilizes an AI-powered scoring system to evaluate prompt effectiveness.
The flow shows how both objective test cases and subjective criteria are merged into a final ranking score.
Sources: apps/nova/messages/en.json:1180-1205
Technical Implementation Details
Nova's backend leverages the Tuturuuu monorepo's shared packages for AI integration and database management.
AI Route Configuration
API routes (e.g., packages/ai/src/chat/google/route.ts) are configured for high performance and streaming:
- maxDuration: 60 seconds.
- preferredRegion:
sin1(Singapore). - Credit Management: Includes pre-flight checks (
checkAiCredits) and post-completion deduction (deductAiCredits).
File Processing
Nova supports analyzing user-uploaded files. The system processes these files from Supabase storage, handles binary data (images/PDFs) as ArrayBuffer, and injects them into the model's message content.
// Example of adding files to AI content parts
// Source: packages/ai/src/chat/google/route.ts:148-185
if (mediaType.startsWith('image/')) {
contentParts.push({
type: 'image',
image: content instanceof ArrayBuffer ? new Uint8Array(content) : content,
mediaType,
});
} else {
contentParts.push({
type: 'file',
data: new Uint8Array(content),
mediaType,
});
}
Summary
Nova is a core component of the Tuturuuu ecosystem, serving as a specialized hub for prompt engineering. By integrating multi-provider AI support, a robust challenge management system, and structured interaction guidelines, it enables both individual learning and large-scale competitions like the NEO League. It functions as the "Conscience" of the platform, ensuring that AI interactions remain aligned, effective, and high-performing.
Sources: conductor/product.md:46, GEMINI.md:15, apps/nova/messages/en.json:670