I build backend systems, web applications, and cloud deployment tools. My work focuses on the parts that make services usable: APIs, billing, deployment recovery, networking, and clear operational workflows.
I'm interested in backend, full-stack, and platform engineering opportunities.
Tools I work with: Python · Ruby · TypeScript · Go · React / Next.js · PostgreSQL · Docker · GitHub Actions
LayerRail — cloud control plane
I develop a Ruby/PostgreSQL cloud control plane built on Ubicloud. My contributions include invoice delivery with retries and duplicate prevention, usage limits, provider integrations, and Kubernetes networking fixes. The repository retains its upstream attribution.
- Explore: Website and architecture.
- Local walkthrough: Docker quick start opens the control plane at
localhost:3000; use Linux, macOS, or WSL. - Validation: RSpec setup and commands. The linked billing change includes tests for retries, transaction rollback, and duplicate delivery.
- Current scope: Active development. Real cloud provisioning requires provider credentials and running workers; opening the local console alone does not exercise those operations.
Meshferry — tunnels from terminal to dashboard
A TypeScript application connecting a CLI agent, Next.js dashboard, PostgreSQL control plane, and WebSocket gateway. It implements terminal approval, subdomain ownership, and HTTP/TCP/UDP forwarding.
- Local walkthrough: Approve a terminal, expose a local HTTP service, then inspect requests and latency in the dashboard. Setup instructions require PostgreSQL/Neon and
DATABASE_URL. - Validation: After building and configuring the database,
npm run smokeexercises approval, HTTP forwarding, and persisted telemetry. Read the test. - Current scope: Development prototype. The smoke test covers HTTP; TCP/UDP behavior and authorization edge cases need broader automated coverage before a production-readiness claim.
My extension of the open-source DevPush platform adds deployment recovery, remote-node management, scoped API tokens, observability, and a Python CLI. These additions are on staging; main contains the upstream baseline.
- Local walkthrough: Follow the Docker Compose quick start, using
git clone --branch staging --single-branch https://github.com/mayowaoladosu/devvpush.git layerrail. Configure a GitHub App and the required secrets before starting. - Validation: Foundation CI passed on July 30, 2026 at
0f6c2e6, covering application/CLI tests, container builds, migrations, BuildKit isolation, and node-agent tests. Local validation commands. - Current scope: Development branch with source and CI evidence. The quick start describes local deployment; hosted and production-scale results are not documented here.
DeployTide — managed deployment platform in development
A deployment platform derived from the Apache-licensed portions of Dokploy. My work includes a framework detector, workspace ownership boundaries, capacity scheduling, and a Go runtime controller.
- Local walkthrough: Development setup uses Node.js 24, pnpm, PostgreSQL, Redis, and Docker.
- Validation: Foundation CI passed on August 4, 2026 at
0248275. Validation notes describe the tested infrastructure components and remaining gaps. - Current scope: Active development, with no public service endpoint advertised. The provisioning templates render infrastructure configuration; real provider lifecycle integration remains future work in the linked validation notes.
Offline Agriculture Advisor — CPU-only AI demonstration
A Python/llama.cpp project using quantized Qwen2.5-3B, static offline field guides, and deterministic output checks for two agriculture prompts.
- Watch: 104-second recorded demo and raw generated answers.
- Reproduce: Model download and profiler instructions.
- Measured results: Committed benchmark output reports 13.41 tokens/sec, 3,458.86 MB peak RSS, and 0.80 ARC-Easy acc_norm across 50 samples on a GitHub Actions CPU runner.
- Current scope: A bounded demonstration. Those results do not establish agricultural accuracy beyond the examples or performance on every laptop; there is no fine-tuning or general retrieval pipeline.
Cloud Steward incident callback — a Python contribution merged into CALL-E's agent examples on September 11, 2026, with tests using a simulated provider and explicit confirmation boundaries for incident notifications.
Evidence last checked September 19, 2026. CI results apply to the linked commits. Local setup guides and tests are provided for reproduction; only explicitly linked CI runs and benchmark artifacts are reported here as completed validation.



