Machine Learning Engineer / Data Scientist

Joaquin Ramirez

I've spent the last three years building GenAI products and predictive models that made it to production: an AI copilot used by 1300+ organizations, fraud detection for a network of 35 banks, forecasting and optimization for exporters and retailers. Now an MSBA candidate at UCLA Anderson.

Los Angeles, CA

Selected work

AI copilot and agents for HR teams

Crehana (HR-tech SaaS) · Senior Machine Learning Engineer · Feb 2025 – present

I lead development of Crehana's GenAI copilot and the agentic systems behind it (LangChain, LangGraph), now used by 1300+ organizations and 48K+ people across Latin America. When our data-extraction agent kept failing in production, I redesigned its routing and added reliability safeguards: failures dropped 25% and adoption of the feature grew 9x. I also present tradeoffs to our executive committee and coordinate the work of two other ML engineers.

1300+ organizations 48K+ users failures −25% adoption 9x

LangChain · LangGraph · OpenAI · Python · AWS

The retrieval pipeline behind the copilot

Crehana · Senior Machine Learning Engineer · Feb 2025 – present

The copilot answers from each company's own documents, so the retrieval layer has to stay fresh and tenant-aware. I built and run the ingestion pipeline: 10K+ documents a week, indexed on a schedule into a search corpus of 1.2M+ vectors on Qdrant with OpenAI embeddings.

10K+ docs/week 1.2M+ vectors

RAG · Qdrant · OpenAI embeddings · Python

Contract allocation for a commodity exporter

Canvia (IT consulting) · Data Scientist · Dec 2023 – Dec 2024

A fishmeal exporter planned its contract allocation by hand, two weeks per cycle. I built the product classification and allocation algorithm that now does it in under 20 minutes, raising estimated gross margin by ~55% and cutting operating costs by ~40%. I presented the results to about 15 client stakeholders, including VPs.

est. gross margin +55% costs −40% 2 weeks → <20 min

Python · scikit-learn · optimization

Fraud detection across 35 financial institutions

Canvia · Data Scientist · Dec 2023 – Dec 2024

Real-time anomaly detection (Isolation Forest) on transactions moving between 35 of Peru's leading financial institutions. The system automatically flags the highest-risk ~2% for human review instead of asking analysts to watch everything.

35 institutions ~2% flagged, real time

Isolation Forest · anomaly detection · Python

G.E.N.I.E., my undergraduate thesis

UTEC · with Alexandra Shulca · 2023

A generative chatbot specialized in our university's regulations, built back when that meant doing everything yourself. We scraped and structured the documents, generated a fine-tuning dataset, and LoRA fine-tuned Falcon-7B through 20 hyperparameter trials, taking evaluation loss down 84% (2.92 to 0.46) and perplexity from 18.5 to 1.6, then validated the answers with user testing.

eval loss −84% perplexity 18.5 → 1.6

Falcon-7B · LoRA · Hugging Face · code →

Skills

AI & ML
LLMs and GenAI (OpenAI, Claude, Gemini) · agentic systems (LangChain, LangGraph) · RAG · vector databases (Qdrant) · scikit-learn · NLP
Analytics
A/B testing · hypothesis testing · forecasting · anomaly detection · classification
Programming
Python (pandas, NumPy) · SQL · R · C++
Tools & Cloud
AWS · Azure · GCP · Docker · Git · Tableau · Power BI
Languages
Spanish (native) · English (fluent) · German (B2)

Education

UCLA Anderson School of Management

Master of Science in Business Analytics (MSBA), expected December 2027 · Anderson Merit Fellowship

Universidad de Ingeniería y Tecnología (UTEC)

B.S. Computer Science, 2024 · Top 10% of class, Academic Excellence Scholarship · Student Body President (2022) · Exchange year at Hochschule Luzern, Switzerland