AI Engineer & Data Scientist with 4+ years building production ML across GCP, AWS and Azure. Right now I build
> multi-agent NL-to-SQL systems.> multi-agent NL-to-SQL systems.> multi-agent NL-to-SQL systems.> multi-agent NL-to-SQL systems.> multi-agent NL-to-SQL systems.> multi-agent NL-to-SQL systems.> multi-agent NL-to-SQL systems.> multi-agent NL-to-SQL systems.
years shipping ML and AI to production
MRR lift with hybrid search + RRF
revenue increase from a hybrid recommender
less time spent on data analysis
interactions behind a production recommender
Every element here has been used in production or published work. Filter by family to see where each one fits.
Measured outcomes from shipped systems — search quality, revenue, latency and hours given back to teams.
Vector + lexical search on OpenSearch Serverless, evaluated with an LLM-as-judge framework (NDCG, MRR, MAP, Hit@k). Image search reached MRR 1.0 in evaluated scenarios.
Neural networks + XGBoost trained on 10M+ interactions, with FAISS vector search for semantic job matching.
Zippia · Recruiting platformConversational database interface with GPT-4o and Claude, dynamic schema injection and SQL guardrails at ~3s latency.
ZamsRAG + Qdrant matching engine aligning Limited Partners with Emerging Managers.
ZamsGCP pipeline for postal routing with HDBSCAN, cutting processing time by 65%.
ZamsNatural-language questions become analysis-ready data and charts, with scheduled queries and Datadog observability.
Amaris Consulting → DecathlonTyped blocks, auto-correction after rejection and a pt-BR number validator. Validated on 10 reports with production data, delivered by idempotent email with retry and backoff.
Amaris Consulting · OPECO agentPyTorch/TensorFlow models for disease classification on X-ray, CT and MRI, with Grad-CAM explainability.
Harpia HealthMongoDB and Elasticsearch index optimization, query rewriting and caching over millions of daily events.
ZippiaIFMBE Proceedings (CLAIB & CBEB 2022): a mobile ML app for hemifacial spasm, ~95% of blinks detected.
UNIFESP · ResearchFrom marketing BI to multi-agent systems, remote with teams in Brazil, the US and Europe.
Leading conversational and multi-agent AI systems that democratize the corporate data lake and automate metrics and insights.
Built the search and data foundation for an AI shopping agent in fashion e-commerce.
Conversational AI, semantic search, production ML and document processing to streamline operations.
Recommendation systems, large-scale data pipelines and GenAI for a recruiting platform with millions of daily events.
Medical imaging pipelines and deep learning for disease classification and anomaly detection.
A unified warehouse for multichannel marketing campaigns.
Products I co-founded and experiments I can't stop tinkering with.
A pre-launch SaaS that imports résumés, scores them against a specific job description and rewrites only the fields that fail review — without inventing experience.
Repositioned a wedding website into a full digital studio: visual identity, monogram, invitations, Instagram posts, WhatsApp communication and the published site. Built the first creative-assets generator.
An AI bouldering partner that sees a wall, reads the climbing on it and coaches the climber. Upload a wall photo, segment the holds, see the overlay.
One hotkey screenshots the focused window, OCRs it with an LLM, translates to Spanish or Portuguese and reads it aloud. Runs on Windows and macOS.
A master’s in progress in Germany, a science & technology degree from UNIFESP and a peer-reviewed paper along the way.
First author · IFMBE Proceedings (CLAIB & CBEB 2022), with Escola Paulista de Medicina. A phone video replaces manual blink counting and infrared gear.
Open to senior AI / ML engineering roles, remote or hybrid. I reply fast.