projects
Each ships only when its case study is published. The write-up — the reasoning, the rejected alternatives, what failed, the numbers — is the product.
A Model Context Protocol server that lets any AI agent build, deploy, observe, and visually verify Android apps on real devices and emulators.
durable skill Designing tools agents can actually use — granularity, recoverable errors, safety gating, output budgets.
Kotlin/JVM · MCP Kotlin SDK · ADB · Gradle
case study coming soon
A Kotlin Android app routing inference between an on-device small model and cloud frontier models using explicit, measured routing logic. The benchmark report is the real deliverable.
durable skill Cost / latency / quality / privacy tradeoff engineering, with real-device numbers.
Android · Kotlin · Compose · on-device runtime + cloud
case study coming soon
A production-grade agent monitoring shipments across Colombian courier APIs, detecting anomalies, explaining them, and proposing or taking actions behind human approval.
durable skill The complete production loop — build → evaluate → observe → iterate — on a framework-agnostic, JVM-native agent core.
Kotlin · Ktor · pgvector RAG · eval harness · OpenTelemetry · Compose client
case study coming soon