Consti Ertel

Building toward the axioms

Selected work

Supply Chain RL Environment

I built SupplyChainBench, an RL environment and benchmark designed to test whether LLMs can learn to make better decisions when the consequences of their actions only become visible several steps later. Using the Beer Distribution Game, models must forecast demand, account for orders already in the pipeline, and make decisions whose effects unfold over several weeks. On 16 held-out stochastic benchmark seeds, Qwen3.5-4B scored 38.42 with a GRPO + LoRA adapter versus 11.78 untrained—a 3.3× improvement. The adapter was trained on the earlier environment and evaluated on the stochastic version without retraining.

RAID: Retrieval-Augmented World Models

A cross-attention mechanism that infers robot actions from world-model state transitions, retrieving the closest demonstrations from memory. Cuts action-prediction error 6.5x over a direct MLP baseline on LIBERO-Spatial using just 25 demos, then fine-tuned with GRPO reinforcement learning.

Fincast

AI-assisted valuation for public equities. Enter a ticker, pick DCF or an exit multiple, and get five-year projections, a fair value, and upside to the current price.

Ideas