Leo Torres

Leo Torres

Senior Data Engineer

LLMs & AI • Python • BigQuery • Leipzig/Berlin
Download Resume Full technical experience
Schedule Call Let's discuss your project

About

I'm a Lead Data Scientist at FGS Global, building LLM-powered data pipelines that process 10M+ documents daily for Fortune 500 clients. Based in Leipzig, Germany (commuting to Berlin), I specialize in large-scale AI systems, distributed data processing, and production data infrastructure.

I architect data infrastructure handling billion-row datasets with sub-second query performance using BigQuery, Python, and modern cloud technologies. As technical lead for a team of 6 engineers on our flagship internal product, I implement cutting-edge RAG architectures and vector databases.

My approach combines rigorous computer science fundamentals with practical engineering solutions. I deliver measurable impact at scale through robust pipelines and APIs that serve real production workloads.

Technical Expertise

Core Competencies

Leading data science teams while building production AI systems. Expert in LLM applications and large-scale data processing, from prototype to billion-row production systems.

LLMs & AI Python (Expert) BigQuery FastAPI Vue.js RAG Systems Vector Databases Team Leadership

Featured Projects

hns — Headless Multi-Agent Orchestration (private)

Headless dispatcher that runs Claude Code workers in parallel through a state machine: PR triage, CI gating, stuck-detection with escalation, orphan recovery, autonomous worktree dispatch.

Design: Python; abstractions and control loop designed from first principles.

Python • State Machines • Concurrency • Git Worktrees

The Aris Program

Founder of a multi-project open-source organization building web-native infrastructure for academic publishing. Designed RSM (Readable Science Markup), a markup language for scientific documents, with a tree-sitter grammar.

Built Scroll Press: a preprint server (FastAPI, PostgreSQL, HTMX) accepting Typst, Quarto, MyST, and Jupyter; live in beta. Architecting a collaborative editor (Vue) as the reference RSM implementation.

Python • FastAPI • PostgreSQL • HTMX • Vue.js • tree-sitter

Website GitHub

XGI: Complex Group Interactions

Co-Lead Developer of Python library for analyzing higher-order networks and hypergraphs. NumFOCUS affiliated project with growing academic user base. Implemented core algorithms, designed API, and established comprehensive testing framework.

Technical leadership: OOP design, CI/CD with GitHub Actions, performance optimization with NumPy and Numba.

Python • NumPy • pandas • Numba • pytest • GitHub Actions • OOP

Documentation GitHub

Manim Community

Organization Owner and Core Developer of the community-maintained version of 3Blue1Brown's mathematical animation engine. Contributing to the open-source Python library that creates precise, programmatic mathematical visualizations and educational content.

Recognition: Featured in GitHub's Popular Python Repositories. Contributions: Algorithm implementations, performance optimizations, documentation improvements, and community support for mathematical animation workflows.

Python • Mathematical Visualization • OpenGL • Cairo • Community Development

Website GitHub

FGS Global — Flagship Intelligence Product

Lead Data Scientist and technical lead for a team of 6 engineers. Designed, built, and own end-to-end the news media analysis pipeline — an LLM-powered system processing 10M+ documents/day into client-facing intelligence for Fortune 500 firms. Architected retrieval infrastructure over billion-row datasets in BigQuery with sub-second query performance.

Design: RAG and vector-search layer made domain-specific retrieval accurate enough for production client use; FastAPI service layer serves multiple internal products from a shared data backend.

Python • FastAPI • BigQuery • GCP • LLMs • RAG • Vector Databases

Data Berlin Meetup Talk

COVID-19 Mobility Data Pipeline

Engineered data pipeline processing mobility data for 300+ US cities during COVID-19 pandemic. Built ETL workflows using Apache Airflow, implemented data quality checks, and optimized geospatial queries with PostGIS.

Results: Enabled epidemiologists to analyze movement patterns in near real-time, contributing to public health policy decisions.

Python • Airflow • Pandas

Professional Experience

Lead Data Scientist → Tech Lead, Data Platform

May 2023 - Present

FGS Global

• Designed, built, and own end-to-end the news media analysis pipeline at FGS — an LLM-powered system processing 10M+ documents/day into client-facing intelligence for Fortune 500 firms

• Architected retrieval infrastructure over billion-row datasets in BigQuery with sub-second query performance

• Designed the RAG and vector-search layer that made domain-specific retrieval accurate enough for production client use

• Technical lead for a team of 6 engineers: set architecture, owned technical direction, drove hiring decisions

• Built the FastAPI service layer serving multiple internal products from a shared data backend

Postdoctoral Fellow

Aug 2021 - May 2023

Max Planck Institute for Mathematics in the Sciences

• Developed algorithms for spectral graph theory applied to complex networks

• Implemented high-performance graph mining tools in Python

• Co-led development of open-source software for the research community

• Published research in leading venues while building practical software tools

Research Intern

May 2019 - Jul 2019

Yahoo! Research

• Built graph representation learning models on Tumblr social network data

• Processed terabyte-scale datasets using PySpark and distributed computing

• Developed Python pipelines for large-scale network analysis

• Applied machine learning to social media data at unprecedented scale

Research Programmer

2012 - 2014

Wolfram Research South America

• Developed data pipelines for the Wolfram|Alpha knowledge engine

• Owned specific data domains end-to-end, including ingestion and quality

• Worked in a remote, globally distributed team

Open Source Leadership

2019 - Present

Multiple Projects

• hns: Headless dispatcher running Claude Code workers in parallel; state machine with PR triage, CI gating, stuck-detection, orphan recovery

• The Aris Program: Founder of multi-project OSS org for academic publishing; designed RSM markup language with tree-sitter grammar; built Scroll Press preprint server

• XGI: Co-Lead Developer, NumFOCUS affiliated Python library for hypergraphs

• Manim Community: Core Developer of mathematical animation library (3Blue1Brown)

• JOSS Reviewer: Regular reviewer for Journal of Open Source Software

Technical Skills

Languages & Frameworks

  • Python (Expert): pandas, NumPy, SciPy, FastAPI, NetworkX, scikit-learn
  • JavaScript (Advanced): Vue.js, HTMX, Node.js, D3.js
  • AI & LLMs: Frontier LLMs (Claude, GPT, open models), RAG systems, vector databases, agentic workflows
  • Other: SQL (advanced), Mathematica, C/C++, R, Julia

Infrastructure & Tools

  • Cloud & Data: GCP (BigQuery, Dataflow, Cloud Run), Azure ML Studio
  • Databases: BigQuery, PostgreSQL, experience with billion-row datasets
  • DevOps: Docker, Airflow, GitHub Actions, Netlify, Fly.io
  • Development: Git, pytest, CI/CD, OOP design, performance optimization

Methodologies & Practices

  • Data Science: Large-scale data processing, ETL pipelines, graph algorithms
  • AI Systems: RAG architecture, vector databases, LLM optimization
  • Development: API design, backend services, open-source leadership
  • Leadership: Team management (6 engineers), technical mentoring, project delivery

Professional Development

  • Speaking: "LLMs: Computation or Measurement?" - Berlin Data Meetup 2024
  • Community: NumFOCUS project leadership, JOSS reviewer
  • Education: Ph.D. Network Science, Recurse Center alum (2016)
  • Languages: English (bilingual), Spanish (native), French (basic), German (basic)