Sr. ML/Mechatronics Engineer — Databricks expert, big data for AI, and scalable production systems.
Building systems that span software, body, and craft. From agent runtimes to breath-hold training, the discipline is the same: define primitives, map failure modes, build recovery paths, iterate toward antifragility.
A Rust-native ecosystem for autonomous software development, built on control theory:
| Layer | Project | Role |
|---|---|---|
| Orchestration | Symphony | Dispatch, lifecycle, and hook management for coding agents |
| Governance | Control Kernel | Typed setpoints, safety shields, multi-rate control loops |
| Improvement | Autoany | Evaluator-Governed Recursive Improvement (EGRI) |
| Kernel | aiOS | Agent OS contract — state, memory, tools, event taxonomy |
| Runtime | Arcan | Runtime implementing the aiOS kernel contract |
| Persistence | Lago | Durable state and event store |
| Stability | Autonomic | Homeostasis controller — operational, cognitive, economic regulation |
| Observability | Vigil | OpenTelemetry-native tracing and GenAI metrics |
| Tooling | Praxis | Canonical tool execution and sandbox engine |
| Networking | Spaces | Distributed agent networking on SpacetimeDB |
Core law: Agent freedom ≤ evaluator reliability.
- broomva.tech — Personal site, AI-native chat, and writing
- chatOS — AI chat app (Next.js 16, AI SDK v6, Better Auth, Drizzle)
- Symphony Forge — CLI scaffolder with composable control metalayer
- Mission Control — Tauri desktop app for multi-project agent orchestration
- Arcan Glass — Dark-first glass design language for web
1. Define the primitive layer — find irreducible components
2. Map failure modes — know where it breaks
3. Build recovery paths — design for when it breaks
4. Iterate toward antifragility — get stronger from stress
5. Make it repeatable — future-self and others can run it
6. Optimize for compounding — will this matter in 10 years?
- "Autonomous Repos Are Control Systems" — The repository is the control surface for autonomous development
- "Harness Over Prompting" — Prompt quality matters, but harness quality keeps agent outputs reliable at scale
- "Autonomy Is Governance" — The autonomy you can grant an agent is proportional to governance quality
Read more at broomva.tech · book.broomva.tech
Mechatronics engineering → industrial automation (PLCs, oil & gas) → data engineering → Databricks & big data for AI → ML engineering → scalable production systems & agent-native infrastructure. Colombian. Freediver. Systems thinker.
Currently AI Lead at Stimulus
14 models and 7 datasets on HuggingFace, including Spanish-to-indigenous-language translation (Wayuunaiki, Nasa Yuwe).
Rust · TypeScript · Python · Databricks · Spark · Next.js · Tauri · Tailwind · Bun · Biome · Drizzle · OpenTelemetry