🧠
RIVALUX
Parallel AI intelligence engine. Simultaneously orchestrates multiple large language model providers, comparing reasoning quality, latency, and output across a unified interface in real-time.
Active
Python
Multi-Model
LLM Research
Overview
RIVALUX is a high-throughput AI benchmarking and racing engine designed to run identical prompts across multiple LLM providers simultaneously. Each provider's response is captured, timed, and scored — enabling direct, side-by-side evaluation of reasoning quality and performance characteristics.
The system is built for research and continuous model evaluation. Results are logged with full metadata per run, enabling statistical analysis of model performance trends over time across different task categories.
Core Capabilities
Parallel Execution
Latency Profiling
Output Comparison
Reasoning Scoring
Task Categorisation
Run Logging
Trend Analysis
Provider Abstraction
Architecture
Dispatch
Async task fan-out layer. A single prompt is dispatched to all configured providers concurrently with independent timeout handling per provider.
Evaluation
Structured scoring layer that evaluates each response on configurable quality dimensions. Produces normalised scores for cross-provider comparison.
Storage
Persistent run log capturing prompt, all provider responses, scores, and timing. Supports replay and offline analysis.
Interface
Live terminal dashboard rendering real-time race results, provider rankings, and aggregate statistics per session.
Key Features
- True parallel model execution — all providers queried simultaneously, not sequentially
- Provider-agnostic abstraction layer — add or swap models without touching core logic
- Configurable evaluation criteria — reasoning depth, factual accuracy, response structure
- Latency and token-efficiency metrics captured per run
- Full run history with replay capability for regression testing
- Aggregated leaderboards across task categories and time windows
Tech Stack
Python
asyncio
LLM APIs
SQLite
Rich (terminal UI)
aiohttp