Applied AI Engineer · LLM Systems · Agentic & RAG Architectures · Local-Model Infrastructure · Quantitative ML
I build and ship production LLM systems end-to-end — agentic orchestration, retrieval-augmented generation, and self-hosted open-model infrastructure — and I pair them with quantitative modeling (Markov chains, Monte Carlo, Bayesian inference, portfolio optimization). I run quantized open models locally (Ollama / Hermes-3, GGUF, VRAM-budgeted inference) and build the reasoning, retrieval, and evaluation layers around them.
My core strength is problem decomposition — taking an ambiguous goal, defining a process, and orchestrating tools and AI to build a working, shipped system. Background: B.S. Mechanical Engineering + M.S. Engineering Management (FIU), robotics, and a decade of multi-domain engineering. Bilingual (EN/ES).
🔭 Currently going deeper: transformer internals & attention mechanics · parameter-efficient fine-tuning (LoRA) · preference alignment (DPO/RLHF) — moving from applied AI toward post-training research.
Python · PyTorch · scikit-learn · XGBoost · PyMC · cvxpy · TypeScript · SQL · Ollama / llama.cpp (GGUF) · Docker · GitHub Actions / CI-CD · Pinecone / pgvector · AWS
| Project | What it is |
|---|---|
| Markov-Wheel | Quantitative options scanner — a 4-state Markov regime chain, a 5,000-path Monte Carlo (GBM) assignment simulator, and Black-Scholes pricing (Greeks, POP, Kelly sizing) must all agree before a trade is recommended. Expected Value is the gatekeeper. |
| hermes-meta-math | Fully local decision engine: a swarm of LLM personas (Ollama/Hermes-3, GPU-resident in 6 GB VRAM) feeds a hierarchical Bayesian conjoint (PyMC); outputs optimized with Markowitz mean-variance (cvxpy) + particle-swarm. Agentic tool-calling intake across 3 Docker services. |
| Taxentia-AI | RAG application for tax research — a retrieval pipeline (embeddings + Pinecone) over a regulated corpus, with automated long-form drafting, ETL, and tiered auth (TypeScript). |
| Axon-Health | AI-native medical-tech platform — federated services with two-tier HIPAA/compliance gatekeepers on a NATS message bus (prototype). |
| Heart-Failure-Prediction | Clinical ML — classification on 12 features for mortality-event prediction, ROC/AUC evaluation. |
| Miami-Housing | Regression ML — feature engineering + gradient boosting for price prediction. |