Diego Ramirez Madriz
I turn ambiguous AI problems into evaluated, usable systems.
The work
Things I've built. Each one you can play with.
Multi-agent reasoning
Do agent teams beat one model at math? Only when the budget allows.
92.0% MAV accuracy, against 79.5% alone
Industry AI agent projectISP support agent
A customer describes a symptom, the agent finds the fault.
47/50 blind test messages read correctly by the model, against 37/50 for rules alone (synthetic)
Industry forecasting projectReady-Mix demand
Hourly concrete demand per plant, so dispatch can plan the busy hours.
37% lower validation error than the historical median
Industry analytics projectFleet stock plan
How many spare parts to stock, and which trucks will need them.
381 mechanical parts modelled
Competition entryFutBot tracker
Robot-soccer video in, robots, ball and possession out.
97.7% of frames with the ball in view found, with no false hits when it's hidden
Course projectOption pricer
Price an option two ways, and see why they agree.
50,000 simulated paths per price
Keep going