Selected work / Full-Stack Developer & System Architect
LLM-HubProMax
- Role
- Full-Stack Developer & System Architect
- Client
- Ali Saad (Self-initiated product)
- Year
- 2026
The core challenge was normalizing providers with different model catalogs, capabilities, rate limits, error behavior, and endpoint coverage behind one predictable OpenAI-style interface. The gateway uses a provider catalog, capability metadata, explicit routing headers, sticky…
Self-hosted OpenAI-compatible API gateway and admin dashboard that routes applications across multiple AI providers through one /v1 interface, with sticky fallbacks, health checks, encrypted local keys, analytics, and multimodal testing.
Three decisions shaped the final experience.
03Unified OpenAI-Compatible /v1 API
One proxy surface covers model listing, streaming and non-streaming chat, tool calling, embeddings, image generation and edits, speech, transcription, translation, and realtime…
Sticky Fallback Routing
Configurable chains keep requests on the selected route when healthy, apply cooldowns after rate limits or provider errors, and expose routed-provider and fallback-attempt metadata in…
Provider Health & Model Discovery
Health checks, model availability discovery, capability tracking, provider ranking, and model sweeps show which routes are configured and currently usable across the provider catalog.
LLM-Hub Pro Max provides one working `/v1` entry point for applications that need to move between models and providers without rewriting client integrations. It supports routed chat, embeddings, images, audio, and realtime session-token workflows, with provider and fallback metadata returned alongside requests. The dashboard centralizes key management, fallback order, model availability, health checks, logs, analytics, diagnostics, and multimodal testing. When a route fails or rate-limits, cooldown and fallback…