Engineering profile
The standards behind the work.
This profile describes how ForgeGuard AI engineers — the practices, stack, and quality bars applied to everything published. It is intentionally about the work; the GitHub organization is the primary record.
Practices
Engineering practices in evidence.
Each practice below is observable in the published repositories rather than aspirational:
API compatibility as a contract
Services implement OpenAI-compatible endpoints so existing SDKs work unchanged; deviations are documented, not discovered.
Health contracts for orchestrators
Liveness and readiness are distinct, warming is reported honestly with Retry-After, and permanent failure exits non-zero instead of lingering.
Deployment determinism
Model weights are baked into images at build time with checksum verification; containers need no network at start; releases are tagged and changelogged.
Security defaults that respect reality
Optional bearer auth via one variable, built-in TLS with sane self-signed defaults for local use, privacy-first logging defaults for PII-bearing workloads.
Documentation as part of the system
Docs live in the repository, versioned with the code, validated in CI, and published to this site through a contract that pins commit SHAs.
Evidence-first publishing
Public claims trace to code, tests, manifests, or releases at a pinned commit. Conflicts between sources surface in CI rather than being silently resolved upward.
Stack
Working surface.
| Languages | Python, TypeScript |
|---|---|
| Serving | FastAPI, Uvicorn, CTranslate2, PyTorch |
| Delivery | Docker, Helm/Kubernetes, GitHub Actions, GHCR |
| Hardware | NVIDIA CUDA (RTX 3000–5000), Jetson Orin edge devices |
| Web | Astro, Starlight, static-first delivery |
Selected work
Read the dossiers.
The project pages are the case studies: each documents the problem, verified capabilities, operating contract, architecture, deployment paths, and limitations.
- ForgeGuard Kokoro Server — OpenAI-compatible local speech synthesis for container and Kubernetes deployments.
- ForgeGuard Faster Whisper Server — OpenAI-compatible local speech recognition with operationally safe container deployment.
- ForgeGuard Hermes Agent — A ForgeGuard-maintained distribution of the Hermes self-improving AI agent, as versioned container images and desktop installers.
- ForgeGuard Markdown Live Preview — A privacy-conscious, browser-only Markdown editor with live preview, packaged as a self-hostable container and Helm chart.
Professional bridge
Talking to the engineer.
The public identity of this lab is deliberately limited to the work itself. For hiring, leadership, or collaboration conversations, use the paths on the contact page.