6+ years experienced AI / Research Engineer building end-to-end AI and data solutions across fintech, healthtech, and e-commerce. Specialized in taking ML models from research to production at scale.
Bachelor of Mathematics, UNAM (GPA: 92/100)
- Staff AI Engineer @ VerveMarket. Working directly with the CTO to own AI/Data strategy and infrastructure. Leading hybrid search, a customer-facing autonomous agent that acts on the live catalog and cart, LLM fine-tuning, multimodal image generation pipelines, and evaluation infrastructure (deterministic, statistical, LLM-as-judge) for batch and real-time serving.
- Previously: Lead AI Engineer @ Contpaqi | Lead R&D @ SoftServe | Sr ML Engineer @ Nubank | Sr Data Engineer @ Citigroup | AI Engineer @ Botco AI | Sr ML Engineer @ BBVA AI Factory
- Hybrid search at VerveMarket driving a 27% add-to-cart conversion lift at ~100K requests/day
- Real-time underwriting risk models at Nubank (Top LATAM Fintech), 3M+ monthly credit decisions on K8s
- 4+ PB Data Lake migration at Citigroup (S3, Snowflake, Spark on EKS)
- Agentic AI platforms for healthcare (SoftServe) and accounting (Contpaqi)
- RAG pipelines and LLM fine-tuning (GPT, LLaMA) at Botco AI, 1M+ documents indexed
- Anti-fraud graph ML models at BBVA AI Factory
Languages: Python, Scala, SQL, Golang, Rust (familiar)
AI/ML & Agents: PyTorch, TensorFlow, HuggingFace, LangGraph, LlamaIndex, CrewAI, MCP, RAG
Serving & Optimization: vLLM, TGI, Ollama, MLX, Apple Neural Engine, quantization (GGUF, MLX), LoRA / QLoRA, Unsloth
Training Data: synthetic dataset generation, distillation, SFT curation, quality gating
Evaluation: LLM-as-judge, significance testing, Arize, LangSmith, OpenTelemetry
Data: Spark, Airflow, Kafka, Hadoop, dbt, Databricks
Cloud: AWS (S3, EKS, SageMaker, Glue, Lambda), GCP (BigQuery, VertexAI), Azure (AZ-900)
Infra: Docker, Kubernetes, FastAPI, PostgreSQL/pgvector, MongoDB, Cassandra, Firebase, Snowflake, Pinecone
- TensorFlow Developer Certificate
- Apache Airflow Fundamentals
- Microsoft Azure Fundamentals (AZ-900)
- Deep Learning Specialization
- MLOps Specialization (Duke University)
- TOEFL (C1 English)
| Project | Area | What it is |
|---|---|---|
| Bespoke-Stratos-ES | Dataset / distillation | 16K-row Spanish reasoning dataset for SFT, distilled from DeepSeek V4 Flash. Traces are generated natively in Spanish rather than machine-translated, and gated on language dominance and truncation |
| es-reasoning adapters | Fine-tuning | LoRA and QLoRA adapters that make Qwen3-4B, Qwen3-1.7B and Gemma-3-4B reason in Spanish, trained on the dataset above and published for MLX, PEFT and GGUF. Measured on a held-out Spanish GSM8K split with paired McNemar tests: Qwen3-4B gains 7.4 accuracy points alongside perfect Spanish (p = 0.0031), Gemma-3-4B gains 27.9 and a thinking mode it did not have, Qwen3-1.7B trades 5.7 points for it and its model card says so |
| mlx-community | Quantization / Code | First MLX port of an unsupported architecture (NVIDIA's Nemotron-Parse OCR), with 4/8-bit quantizations published with measured OCR fidelity. |
| langchain-searchapi · llama-index-tools-searchapi | Agent tooling | Two standalone PyPI packages bringing multi-engine web search to LangChain and LlamaIndex |
Links go to the pull requests.
| Project | Area | Status |
|---|---|---|
| ANEForge | Apple Neural Engine | Ongoing contributor |
| mlx | Apple Silicon inference | Contributor |
| mlx-lm | Apple Silicon inference | Contributor |
| mlx-vlm | Vision-language models | Contributor |
| vLLM GGUF plugin | LLM serving | Contributor |
| OpenInference | LLM observability | Contributor |
| inspect_ai | Model evaluation | Contributor |
| Qwen Code | Coding agent | Contributor |
| Harbor | Agent framework | Contributor |
| scrapy-playwright | Scraping | Contributor |
| LangChain Docs | Documentation | Contributor |
| coremltools | Apple Neural Engine | In review |
| tokenizers | Tokenization / Rust | In review |
| LiteLLM | LLM gateway | In review |
| langchain-google | LangChain | In review |
| smolagents | Agent framework | In review |
| CrewAI | Agent framework | In review |
| MCP Servers | Agent tooling | In review |
| Project | Description |
|---|---|
| LLM Chatbot | Generative QA with Vector DB (Pinecone) + Streamlit |
| AI Disease Diagnoser | Symptom analysis with LLMs on GCP serverless |
| GEC System | Grammar Error Correction with T5 and OpenAI |
| CONAGUA ELT | Weather data pipeline (Airflow, Docker, S3, BigQuery) |
| CONAGUA Scala | Same pipeline in Scala/Spark |
- English (C1, TOEFL) | Spanish (Native) | Portuguese (Learning)