R.A.T.E. Intelligence

Reinforcement-based Assessment of Target Elements — A deterministic ML platform that fuses Deep RL with ANOVA statistics to identify the optimal predictive factors in any dataset.

LLM-as-Prior

Groq Llama-3 analyzes metadata to propose targets & features. Its suggestions are injected as initial weights into the RL agent — not as decisions.

Deterministic Execution

All ML computations are seeded (random_state=42). Same data in = same results out. Every single time. No statistical drift.

Hallucination Gate

A strict validation barrier cross-checks every LLM suggestion against actual CSV headers. Hallucinated columns are instantly purged.

Chunked Uploads

Files are split into 5 MB chunks and streamed to the backend. Supports datasets up to 100 MBwith a flat memory footprint.

Redis Session Cache

Duplicate analyses return instantly via Upstash Redis. Results are cached for 1 hour, then self-destruct automatically.

Disposable Architecture

Zero permanent storage. Raw CSVs are auto-purged after analysis. MongoDB stores only tiny metadata receipts with a 1-hour TTL.

Next.jsFastAPIPPO (Stable-Baselines3)ANOVAGroq Llama-3RedisMongoDB