AI & Machine Learning Engineer
Engineering notes, empirical studies, and systems built with evidence, evaluation, and operational honesty.
Interactive reliability check
System anatomySelected checkpoint
Can the system trace every important claim to a trustworthy source?
Inspect provenance, coverage, freshness, and split integrity before asking a model to learn or retrieve anything.
Design principleBad evidence becomes confident error.
Enter the full Reliability LabWriting
Empirical studies, visual field guides, failure analysis, and reflections on learning through a technical transition.
Empirical study
Research essay
Technical essay
Northeastern feature
Systems
Each project includes evaluation evidence, operational limits, and a public implementation record.
Model development and evaluation
A 6.3M-parameter encoder trained from scratch on 37.6K English and Spanish claims, paired with retrieval, RAGAS evaluation, ablations, and a documented limitation audit.
Custom 16K BPE tokenizer, PyTorch training loop, retrieval evaluation, human-evaluation protocol
View repositoryPersistent multi-service AI system
A local AI operating system that connects semantic job matching, MCP-accessible memory, local model fallbacks, document generation, and a unified command center.
FastAPI, MCP, SQLite, MongoDB, ChromaDB, Ollama, TypeScript
Private repositoryProvider-independent orchestration
Four stateless model specialists share graph-backed context through a trace ledger, with graceful degradation, local fallbacks, and 15 hermetic tests.
Python, Pydantic, Neo4j, Claude, ChatGPT, Gemini, Perplexity
View repositoryConversational text-to-SQL research
A hand-annotated multi-turn corpus with 98.6% inter-rater agreement and 88.5% leakage-free held-out execution accuracy after pipeline repair.
NLP, PostgreSQL, corpus annotation, execution evaluation, failure analysis
View repositoryMethod
I care about what a system actually did, what the evaluation measured, and what happens when a provider or assumption fails.
Map the real workflow, decision owner, and adoption barrier.
Verify behavior through telemetry, tests, and traceable state.
Separate benchmark wins from leakage, bias, and test gaps.
Design fallbacks, recovery paths, and clear success criteria.
Experience
Before building AI systems, I spent 15 years translating cloud, data, governance, and infrastructure products into adoption across complex enterprise markets.
Manager, Partnership Alliances and Channel Sales, Latin America
Led regional strategy for data intelligence, governance, metadata, privacy, and AI-governance use cases.
Manager, Enterprise Sales and Channel, Latin America
Built a zero-to-one regional market, scaled revenue 5x, and earned Global Salesperson of the Year.
Business Development Consultant, Partner Solutions
Scoped cloud, database, middleware, and platform solutions while building a $20M+ enterprise pipeline.
Enterprise pricing, partner support, and regional growth
Built the commercial judgment behind pricing, adoption, analytics, and high-trust customer delivery.
About the author
I connect technical systems, customer reality, and executive decisions.
My background spans AI evaluation, agentic systems, enterprise cloud, data governance, partner ecosystems, and zero-to-one regional growth. That combination is especially useful when the problem is ambiguous and the system must earn trust in practice.
Based in Greater Boston. Bilingual in English and Spanish. Open to on-site delivery and frequent travel.
M.S. Data Analytics Engineering, August 2026
GPA 3.7
B.S. Economics
Amazon Web Services
Spanish AI Data Trainer, Alignerr by Labelbox
Greater Boston and open to travel
Available for AI engineering, research engineering, and forward-deployed work.