0
Downtime migrating the core ML scheduler to regional federated compute across thousands of pipelines.
SWE/MLE · ML Systems
Felipe Felix Arias
SWE/MLE on Uber’s ML Training team. I build the systems that run large GPU training workloads and keep them reliable, efficient, and cost-aware.
Founder of Marovi AI, a wiki-style translation platform that makes research and learning materials accessible across languages. Former NSF Graduate Research Fellow at UIUC.
ML Infrastructure
GPU Training
Kubernetes
Ray
Agentic Solutions
Python
Go
Bay Area based · Fluent in English and Spanish
Impact at a Glance
Signals from production systems, research, and Marovi AI.
3.7%
Reduction in unsafe behavior from a multi-signal distracted driving system.
Top 10% x2
Y Combinator selections for Marovi AI.
5x
Faster multi-agent motion planning via spatial regression models.
Technical Focus
Systems areas I work in.
Scalable Training Systems
Infrastructure for large GPU fleets: reliability, throughput, and orchestration.
GenAI Systems
Production GenAI: assistants, safety tooling, and evaluation loops.
Machine & Agentic Translation
Translation infrastructure for research and learning materials across languages.
Fine-Tuning
Fine-tuning strategy, partnerships, and production rollout.
Experience Highlights
Selected work across ML infrastructure, GenAI, and translation.
Uber — ML Training Team
- Build job controllers and GPU training infrastructure; led the move from zonal to regional federated compute with zero downtime.
- Co-designed Ray job submission across federated Kubernetes clusters and shipped AI safety systems plus internal assistants for incident response.
- Leading mTLS rollout for batch workloads and serving as primary point of contact for LLM fine-tuning partnerships.
Marovi AI
- Built Marovi AI Wiki, a translation platform that makes research and learning materials accessible across languages.
- Designed correction loops that blend user edits, reviewer votes, and canonical translations; selected top 10% at Y Combinator twice.
- Implemented cost controls and enterprise-ready deployment options.
Research — UIUC Parasol Lab
- Built spatial regression models that improved multi-agent motion planning speed and collision avoidance.
- Released tools for indoor multi-agent navigation and dataset analysis.
- Master’s thesis on motion pattern prediction; earlier work at Uber Search (2023) and Stanford Hazy Research (2018–2019).
Selected Projects
A few representative projects.
Marovi AI: Translation Infrastructure
Wiki-style translation platform that makes research and learning materials accessible across languages, with verification and quality feedback loops.
Avoidance Critical Probabilistic Roadmaps
Learns avoidance-critical regions to build roadmaps that scale to dynamic, multi-agent motion planning.
Multi-Perspective Navigation with RL
Uses multiple camera perspectives to improve navigation policy learning.
Video as a Sensor for Urban Risk
Weak supervision and behavioral heuristics surface risky roadway events when labeled data is scarce.
Weakly-Supervised Multi-Sentence Relation Extraction
Weak supervision scales relation extraction beyond single sentences.
Publications
Selected academic work.
Now
Focused on ML training infrastructure at Uber, plus safety tooling and fine-tuning partnerships.
Outside Work
I make music and shoot photos/video. Highlights on Instagram.