Lorenzo Orozco

Engineer specializing in full-stack AI: LLM systems and machine learning.

I architect multi-agent frameworks, RAG systems, evaluation loops, and production pipelines, combining generative AI with solid ML foundations to deliver solutions that survive real-world complexity.

View my work experience Explore my projects

What I build

AI systems that blend engineering rigor, thoughtful evaluation, and product sense.

LLM agents and evaluation

RAG pipelines, multi-agent flows, and guardrails that hold up under real usage.

  • Retrieval, grounding, and safety layers with reproducible evaluation.
  • Tracing and observability to surface drift, hallucinations, and latency.
  • Prompt and tool design tuned to business outcomes, not demos.
LLMs RAG Evaluation LangGraph Observability

Data, infra, and automation

Reliable plumbing for models and products: ETL, orchestration, and deploys.

  • APIs and services in Python/FastAPI with structured logging and metrics.
  • Feature pipelines, data validation, and backfills for trustworthy inputs.
  • CI/CD, environment parity, and playbooks that keep releases calm.
Python FastAPI Pipelines CI/CD Automation

Product craft

Translate vague requests into scoped features, prototypes, and shipped outcomes.

  • Discovery with stakeholders; define success metrics and experiment plans.
  • Design lightweight UIs and flows that make ML output usable and trustworthy.
  • Documentation and handoffs that keep teams aligned after delivery.
Product UX Delivery Experimentation

Recent highlights

Snapshots from production work and research that shaped how I build.

Internal tools coordinator

L'Oreal | 2026 - Present

  • Squad lead for building in-house platforms to replace vendor tools with tighter control and integration.
  • Designed APIs, data flows, and dashboards for marketing, operations, and business teams.
  • Set standards and reusable components so new tools launch faster and stay maintainable.

AI engineer

Monico | 2025

  • Architected enterprise RAG pipelines; question extractor reached 92% precision on complex legal docs.
  • Hybrid retrieval (embeddings + BM25 + FAISS) improved recall by 30%; modular LangGraph agents.
  • Deployed OCR, ETL, and webhook reporting on Azure with prompt evaluation frameworks.

PINN research for solar cells

Universidad Iberoamericana | 2023 - 2024

  • 4-layer physics-informed network hit MAE 0.00065 and R2 0.92 on absorption predictions.
  • Boosted absorption AUC by 112%; built Flask API and interactive UI for real-time inference.
  • Automated 1,710 curve comparisons and earned a top 5% research scholarship.

Where to start

Pick a path into my work: production systems, research, or side projects.

Work experience

Delivery of AI and software products, from discovery through production operations.

Go to page

Machine Learning with Science

Scientific ML rooted in physics intuition, careful experimentation, and model evaluation.

Go to page

School life

Physics engineering with ML specialization, research projects, and applied builds.

Go to page

Personal projects

Independent builds: AI agents, learning platforms, and tools crafted into working products.

Go to page

Games

Quick, interactive quizzes based on my work and interests.

Go to page

Hobbies

Movement, games, and language learning that keep me balanced and observant.

Go to page

About

Builder mindset, research discipline, and calm delivery.

I stay close to data, instrumentation, and evaluation so models behave in the real world. I enjoy translating messy requirements into scoped work, writing the code, and leaving behind clear documentation and metrics. The goal: systems that earn trust quietly and keep working after the launch post fades.