_{senior mlops · foundations @ baubap · ddd enthusiast}

I'm Juan. I build the plumbing under ML systems — queues, infra, observability, the boring pieces that keep things from breaking at 3 AM.

I love ML infra in general, not just the LLM corner of it. Classical models, GBMs, embeddings, training pipelines, model serving — all of it. The interesting problems usually aren't about the model, they're about the pipeline around it.

I work at Baubap on the Foundations team — the platform layer the rest of the company builds on top of.

I take Domain-Driven Design seriously (not just in slide decks), type-checked Python, and infra you can reason about.

• aiokpl — rewriting AWS's Kinesis Producer Library (which is C++) in pure-Python async. More out of curiosity than necessity
• embeddings at scale — vectorizing large text corpora with Qwen3 on spot GPUs, wrestling with the economics of batching
• baubap-toolkit — internal library so teams stop reinventing anyio + spawn-safe multiprocessing every time
• infra as code — Pulumi on AWS (Batch, ECR, EKS, observability)

• 🇨🇴 Contik — building a platform for Colombian SMBs. Helping the pymes go digital, one workflow at a time
• 🐍 aiokpl — pure-Python async port of the Kinesis Producer Library
• 💬 whatsapp-web.js-service — REST + webhook wrapper around whatsapp-web.js

_{"technology should scale with purpose — not just with code"}