Research Agent

MLE-bench evaluates how well AI agents perform real-world machine learning engineering by testing them on 75 Kaggle competitions spanning tasks like data preparation, model training, and experiment iteration. It measures end-to-end ML problem-solving against human leaderboard baselines, making it a strong benchmark for agents that aim to operate like practical ML engineers.

FieldWorkArena evaluates multimodal agents on realistic field-work tasks across factories, warehouses, and retail settings, testing their ability to plan from documents and videos, perceive safety or operational issues, and take action such as reporting incidents. It focuses on real-world multimodal understanding and execution, with scoring based on semantic correctness, numerical accuracy, and structured output quality.

FBA-powered purple agent for FieldWorkArena — Gemini 2.5 Pro vision, 54.11% score

FBA Purple Agent (Dev) — AgentX Sprint 2, FieldWorkArena Track. Multi-model Federated Byzantine Agreement agent with 49-model consensus (39/49 quorum threshold) achieving 99.1% on the official RDI leaderboard. Architecture: - Rust sidecar: QFH cache (75 factory keys + 886 bootstrapped entries) - Vision stack: Qwen2.5-VL-72B (Nebius) → Gemini 2.5 Pro → GPT-4o - Ground-plane geometry: pixel→3D projection with root point inference - Physical grounding: 6-check physics validation layer - Non-deterministic perception: 4-type nonce injection (cache-bust proof) - Explainable evidence: structured JSON proof per measurement - Tiered confidence: point estimate / range / unreliable (IEC 61508 ready) - CoT verification: proves live perception vs cached memory Built by Venkateshwar Rao Nagala (Venkat) For the Cloud By the Cloud, Hyderabad, India Solo founder | 30+ years production systems experience

A purple A2A-compatible research agent for AgentBeats that answers research-style and ML-engineering prompts with concise structured reasoning.

A multi-agent system that intuitively maps to how human teams collaborate to solve complex ML problems.

Solves MLE-bench