Carlos D. Escobar-Valbuena
Agent OS architect and AI engineering lead. I build AI-native, multi-tenant data platforms in production — across regulated and high-stakes domains. Concurrent senior leadership across Stimulus (AI Lead), TEAM International (Data Architect), and Wedi Pay (Co-founder/CTO, 2024–2026). Open-source author of Life Agent OS, the bstack agentic harness + broomva/skills, Lago, Vigil, and the RCS paper series. MSc AI at Universidad de los Andes (2026).
Live signals
Pulled directly from GitHub and crates.io — what the substrate is actually doing this week.
Updated hourly · 82 ★ · 1.0k crate downloads
Knowledge graph
The Broomva bookkeeping pipeline scores and promotes raw extracts into a queryable entity graph. This is what's in the graph right now — last sync 1mo ago.
Concurrent leadership
Three Principal-tier roles in parallel since late 2024 — compressed surface area across fintech / payments, agentic procurement, and enterprise data platforms.
AI Lead
Architecture redesign of an AI-native, Databricks-backed agentic procurement platform serving FIFA 2026 and the Olympics Committee. Lakehouse-native real-time substrate: Databricks Zerobus streaming into a Unity-Catalog-governed medallion architecture (bronze/silver/gold). Multi-document RAG with policy-bounded tool adapters; multi-tenant JWT hardening.
Co-founder & CTO
Agentic cross-border B2B payment orchestration. Multi-tenant Postgres RLS, Kafka/Redpanda event-driven backbone, Databricks gold-table ETLs. Wedi Agents execute payment actions under strict policy/consent/scoped credentials with full audit trails. Signed and shipped payment-rail integrations — Prometeo (open-banking A2A), TruBit/VelaFi (USDC on/off-ramp), Thirdweb (web3-native) — on a provider-agnostic integration layer.
Data Architect (Contract)
Architectural oversight on the Time Series Engine pipelines I previously led as Sr. ML/AI Tech Lead. Anchor customer: a NYSE-listed US natural-gas operator with ~70K wells across multiple US basins. Multi-tenant streaming + batch telemetry on Databricks + MLflow + PyTorch handling 10K+ records/sec. Production medallion-style bronze/silver/gold tables since 2021.
Selected engagements
Anonymized references — case studies available under mutual NDA.
Signed and shipped integrations with Prometeo (open-banking account-to-account), TruBit — now VelaFi (USDC on/off-ramp), and Thirdweb (web3-native) behind one provider-agnostic layer at Wedi Pay.
Anchor customer of the Time Series Engine. Production medallion-style lakehouse since 2021 with anomaly detection, alarm threshold estimation, and digital-twin integration.
Stimulus customer surface — agentic supplier intelligence with policy-bounded tool adapters and multi-document retrieval.
Production-grade agentic RAG over an internal SageMaker engineering knowledge base — pgvector + FAISS, three switchable interaction modes (graph-based, tools agent, conversational). Delivery accepted at evaluation stage.
Polars-based ML pipeline integrating prints/taps/payments event streams into a feature-engineered training set; GridSearchCV tuning with full MLflow autolog reproducibility.
Open source & research
The substrate behind everything else — published as MIT/Apache OSS.
Most-starred repos
Live from GitHub. Updated hourly.
Life Agent OS — published crates
Live from crates.io. The Rust substrate of the agent runtime.
Recent writing
Long-form essays and short notes from the build.
Writing
Notes
Stack
Lakehouse & Data Platforms
AI / RAG / Agentic
Cloud & Observability
Languages & Runtime
Elsewhere
Download CV
Three lengths — pick the one that fits your read time.
Talk to me
If you're building production agent systems, multi-tenant lakehouses, or governed AI platforms — and want to compare architectures or discuss a role — open the contact page with your specific bottleneck. Detailed engagement narratives available under mutual NDA.