Jonathan Siles

• Built production AI workflows for recruiting, marketing, and affiliate review: LLM-based applicant shortlisting, competitor intelligence over a 7,484-ad corpus, and automated partnership scoring. • Built an AI creative engine letting non-designers produce brand-aligned marketing assets: LLMs handle creative direction, deterministic code handles layout (text overflow, contrast, face occlusion across export formats).

• Architected and operate a real-money algorithmic trading system in Python: 7 concurrent ML scanners across 4 books, ~60k decisions in six months, longest unbroken run 3 months, tail latency cut from seconds to milliseconds. • Hand-rolled a 14-feature online-retraining classifier; rejected an XGBoost model that won offline AUC in 6/6 folds but cut realized edge ~40%; changes ship only on paired A/B.