Diego Llosa Tejada

I build the AI company-generation system behind the private equity simulator at Drexel's Stamatakis Center for Alternative Investments, working with Dean Emeritus George Tsetsekos. It produces the CIMs, teasers, and financial statements that Vanguard MBA students run diligence against. Each company is generated with deliberate flaws in its projections, its customer base, and its capital structure, so that diligence has something to find.

I am now building the macro layer: an engine that takes a shock, say a five percent rise in the minimum wage, and pushes it through every company differently depending on sector and archetype. A turnaround runs on thin margins and no slack, so that same wage increase moves its revenue and EBITDA far harder than it moves a steady-state company's. A healthcare platform compounds faster off the same demand shock than a consumer business does. Every sector needs its own elasticities, derived and defensible, before the following year's financials can be regenerated from them.

The Alignment Paradox is an iterated prisoner's dilemma tournament across four language models, testing whether the newer ones cooperate better than the models they replaced. They don't. The newer models defect from the first round and settle into mutual punishment, while their predecessor held cooperation to the endgame and scored roughly twice as well.

Presented at the 4th Annual PREDOC Conference at Chicago Booth. Posted on SSRN with the payoff matrix, model versions, and prompts published for replication.

At sixteen I co-founded a venture importing American confectionery into Arequipa, the first in the city to carry Feastables. The launch ran as a one-day drop in which each bar's price rose with every unit sold. Inventory cleared in a single afternoon.