Eric Agyemang DevSecOps & Platform Engineer

Use case

A portfolio should prove the craft, not just list it. So this site ships itself the way a production frontend ships: no manual uploads, no click-ops.

Architecture

• Static Astro build (zero client JS by default) + Tailwind v4 design tokens.
• GitHub Actions pipeline: install → type-check → build → deploy on every push; a unique preview URL per pull request.
• Served from Cloudflare Pages' global edge with automatic HTTPS and cache invalidation.
• Motion (Framer Motion) islands for scroll-linked animation, reduced-motion and no-JS safe.

Use case

MCP is becoming the standard way AI clients call external tools, and every new server rewrites the same scaffolding. This is the boring-but-critical 80% done right, so the next one starts at mile 10.

Architecture

• High-level MCP SDK with zod-validated tool schemas: one definition drives validation, typing, and tool discovery.
• Construction is separated from transport, so the same server is tested over an in-memory transport with no mocks.
• Errors contained per call; stdout reserved for the protocol, logs to stderr.
• Multi-stage, non-root Docker image + CI (lint, type-check, test, build).

Use case

A leaked key in git history is irreversible: public commits are scraped within minutes and the only fix is rotation. leakguard catches secrets at commit time, when they're still cheap to fix.

Architecture

• Three detection layers: 90 provider rules + Shannon-entropy detection for unknown secrets + sensitive-filename rules.
• Runs as a CLI, a git pre-commit/pre-push hook, and a GitHub Action, for defense at every stage.
• Entropy false positives tamed with guards (hashes, integrity sums, PEM/public keys, base64-decode check).
• Ruleset built via a generate → adversarial-corpus → gap-audit loop; verified by per-rule + corpus tests (43 real secrets flag, 39 decoys stay clean). Zero runtime dependencies.

Use case

Apps that depend on a single AI provider break when it has an outage, rate-limits, or a price spike. This routes across providers so the app stays up and costs stay visible.

Architecture

• Redundancy: provider-agnostic adapters (OpenAI / Anthropic / OpenRouter over fetch) in an ordered fallback chain, so no single provider is a point of failure.
• Resilience: retries (backoff + jitter), per-provider circuit breaking, and per-attempt + terminal overall deadlines.
• Idempotency: an idempotency key coalesces concurrent identical calls into one upstream request, so a retry never double-charges.
• Observability: lifecycle events, a pull-based metrics() snapshot, and a per-call audit trail. Adversarially audited; 24-test failure-mode matrix; zero runtime deps.

Use case

Manually editing a long video into vertical Reels/Shorts with captions is hours of repetitive work. This automates the reframe, format, and caption burn-in.

Architecture

• Deterministic FFmpeg graph: reframe, overlay, and burn captions; H.264 / yuv420p outputs every platform accepts.
• Containerized so it runs identically on a laptop or in CI.
• Config-driven cut list and branded overlay templates.

Use case

One-off scraping and automation scripts rot fast. This is the durable backbone: typed actions, retries on flake, structured output, and evidence for every run.

Architecture

• Typed page-object actions with retry/backoff on flake.
• Structured JSON results plus screenshot artifacts for every run.
• Headless or headed, and CI-friendly.