Senior / Staff / Principal Software Engineer
Secure systems · Agentic AI infrastructure · Platform engineering · Mobile
micrantha.com · ryanjennin.gs · email
I build long-lived software systems where correctness, security, operability, and explicit trust boundaries matter.
My background spans 15+ years across Android, iOS, React Native, Kotlin Multiplatform, backend systems, cloud/platform engineering, infrastructure, and application security. I still work across the stack, but my current focus is the intersection of agentic AI, secure software delivery, platform engineering, and governance.
The recurring question behind much of my current work is:
How do autonomous systems become more capable without quietly accumulating unchecked authority?
- Governed agentic systems — policy, capabilities, approvals, evidence, provenance, replay, and bounded side effects.
- AI-native SDLC — treating agents as constrained delivery participants rather than trusted autocomplete or autonomous maintainers.
- Context integrity — deterministic prompt/context composition, manifests, lockfiles, provenance, and reproducible agent inputs.
- Agent identity & authorization — sender-constrained OAuth, proof-of-possession, replay resistance, and MCP security boundaries.
- Local-first AI infrastructure — model routing, supervisor/specialist agents, memory, evaluation, scheduling, and secure execution on self-hosted infrastructure.
- Platform & developer experience — paved paths, strong contracts, predictable failure modes, observability, CI/CD, and incremental modernization.
A principle I keep returning to: a system may propose its successor, but it should not independently authorize the transition to that successor. Durable state changes need an external trust boundary and evidence that can be independently evaluated.
| System | Engineering evidence |
|---|---|
| Anthesis and Governance Lab | Governed agent effects, explicit authority boundaries, executable adversarial/conformance trials; runtime enforcement remains integration-dependent. |
| Config Drift Guard | Deterministic control-plane reconciliation, immutable approval plans, stale-state rejection, SQLite workflow persistence, and an operator UI. |
| Repora | Go repository control, explicit plans, safe ref reconciliation, and durable execution evidence (experimental). |
| Eyespie | Kotlin Multiplatform offline-first mobile computer vision, native Android/iOS integration, signed game interchange, and exact-candidate release verification; physical acceptance is still open. |
| TidyFS | Rust filesystem analysis and reversible cleanup, recovery and adversarial validation, with AI limited to advisory decisions. |
| Amaryllis | React Native/on-device multimodal inference and schema-constrained component integration. |
Additional work: Hyperion Community (infrastructure/GitOps), Keylix Community (OAuth/DPoP sender binding), Seamwork (modern C++23 interface studies), Shopping Inventory (private data, public implementation and deterministic evidence), and git-autocommit (bounded AI-assisted Git commits).
- Secure Agent Workflows — secure AI-assisted delivery patterns, task boundaries, threat modeling, and review evidence.
- Micrantha engineering contracts — shared secure-SDLC, testing, CI/CD, repository and release standards, with no blanket certification implied.
- Personal technical work — music tooling, guitar/DSP experiments and local-first projects.
Historical repositories remain available for context; archived examples do not imply current support or current engineering style.
A set of deliberately separated projects exploring those boundaries:
flowchart LR
C["Calathea<br/>—<br/>workflow"] --> I["Invokrum<br/>—<br/>context identity"]
I --> D["Dubnium<br/>—<br/>bounded execution"]
D --> K["Keylix<br/>—<br/>sender binding"]
K --> A["Anthesis<br/>—<br/>policy · approval · evidence"]
A --> E["Governed effects"]
| Project | Role |
|---|---|
| Anthesis | Deterministic governance boundary for consequential agentic effects: policy, approval, capabilities, evidence, provenance, and auditability. |
| Invokrum | Deterministic prompt-overlay composition and attestation: explicit ordering, compatibility, manifests, lockfiles, and reproducible context. |
| Keylix | Sender-constrained authorization and proof-of-possession primitives for OAuth, DPoP, MCP, and agentic workloads. |
| Dubnium | Private local-first AI/runtime test bed: supervisor/specialists, model routing, memory, tools, scheduling, runtime evidence, and governed execution. |
| Calathea | Workflow/task-contract layer for composing bounded agent work without collapsing orchestration into governance or runtime concerns. |
The goal is not a monolithic agent framework. The interesting work is at the boundaries: what context entered the system, who or what is acting, what authority it has, what effect is requested, who approves it, and what evidence remains afterward.
I tend to work best on systems that are already real enough to have constraints:
- production software with users, migrations, compatibility requirements, and operational history;
- security-sensitive authentication, authorization, credential, and trust boundaries;
- platform or developer-experience work that must improve delivery without creating a second platform nobody wants to maintain;
- mobile systems where platform-specific behavior matters and cannot always be abstracted into "just another web client";
- agentic/AI systems where probabilistic reasoning needs deterministic controls around consequential effects.
I prefer incremental modernization over rewrites, explicit contracts over convention, boring cryptography over clever cryptography, and evidence over claims that a system "should" be safe.
I am continuing to turn the Micrantha work into small, composable, auditable systems and public reference material for secure agentic software delivery.
Professionally, I am interested in Staff / Principal engineering and selective contract work around AI infrastructure, platform engineering, application security, developer experience, secure SDLC, and complex mobile/backend systems — preferably remote-first in Canada.
I am also interested in collaborating on work around:
- agent governance, authorization, observability, and evaluation;
- MCP and agent/tool security;
- secure platform and API design;
- mobile security and device/app trust;
- local/self-hosted AI infrastructure;
- practical research-to-production engineering.
I play guitar and piano, record music, play hockey, and occasionally turn engineering energy toward things that have nothing to do with infrastructure — including Entanglement of Ages, a long-form fiction/screenplay project.
he / him