Hi, I’m Daniela
I design experiments, build causal models, and ship the software that acts on them. I’m finishing a STEM master’s in Quantitative Economics at UCLA, and the projects below are my own builds: AI systems where the judgment is mine and every number comes from tested code.
Selected AI Projects
mlfactory
Point it at any table (a file, Postgres, SQLite) and get a model you can audit. AI agents hunt the leakage that quietly ruins models. A unit-tested CLI computes every number. It caught a planted leak faking a 0.9999 AUC. The honest model scored 0.66.
- Rigor284 tests · sha256 lineage
- AImulti-agent orchestration
churnpilot
Most churn models target who's likeliest to leave. Often those are the ones you can't save. Point it at your customer table and it finds the people an offer will actually change. On simulated A/B data, same $4,500 budget: 72% more value kept. A third as many nudged toward the door.
- Rigorcausal effect, scored on ground truth
- AIagent proposes, the DS decides
Personal Finance Intelligence Agent
Log a purchase the way you'd say it: “$200 pesos en tacos.” It categorizes, converts the currency, and warns you before you blow the budget. The model handles the words. Tested code handles every number, so the math is never hallucinated.
- Shipslive, in daily use
- AIplain language in, tested code out
- Rigorthe model never does the math