Juan C.

I build AI agent systems that run real business operations — not demos. At a US construction-technology marketplace I architected the agent platform behind 25+ production automations. Most engineers can call an LLM API. The hard part is everything around it: concurrency, idempotency, state that survives a restart, and data underneath that you can actually trust. I've spent 6 years on both halves of that problem. ── AI & AUTOMATION ── • Agent platforms — FastAPI services on Kubernetes orchestrating agent harnesses (OpenCode, Claude Agents SDK, Pi) with concurrency locks and autoscaling, plus higher-level abstractions that let non-technical staff define their own workflows. 25+ automations now running across the business. • LLM automation of real workflows — heuristics + LLM triage over SQS/SNS and multi-step AWS Lambda, routing each issue to the right engineer in Jira or Slack. ~400 engineering hours saved per month. • RAG systems — vector search over authoritative documents (Weaviate, OpenAI embeddings) driving real-time suggestions inside an existing workflow. ── DATA PLATFORMS ── • CDC & replication — I architected a company-wide PostgreSQL→BigQuery CDC platform using audit triggers, watermark-based extraction and idempotent partition-bounded MERGE loads. 4+ years in production at a 15-minute freshness SLA, near-zero maintenance. • Warehouse cost control — cut BigQuery spend by $35k/year through partitioning, clustering, pre-aggregated tables and per-query cost attribution, without restricting analyst access. Held for 3+ years. • Pipelines & modeling — Airflow on AWS (1,500-metric pipeline), dbt, star schemas with SCD2, Kafka/Redpanda with Avro and schema-registry contracts, Iceberg, Spark. ── BACKEND ── Python (Django, FastAPI) and Go on AWS and GCP. Event-driven services with the data contracts, idempotency and graceful shutdown that keep them correct under load. ── HEALTHCARE ── I'm also a qualified physician (MD, 2019) before I moved into engineering, which makes me unusually effective on clinical software: I've built a multi-tenant practice-management system for a multi-clinic gynecology practice (Django, React, PostgreSQL), and won first place at the Saptiva AI Hackathon for a RAG system that suggests evaluations in real time as a doctor completes a medical record. If your project touches clinical workflows, you won't have to explain the domain to me. Background: MSc Data Science (Summa Cum Laude), MIT MicroMasters in Statistics and Data Science, MD. I currently lead a platform team of 6 engineers. Open source: author of ToolEngrams, with merged fixes in OpenCode and Rapid-MLX. If you have an internal process that's slow, manual, or quietly eating your team's week or a data platform that keeps breaking, tell me what it looks like and I'll tell you straight whether it's worth building.