←SELECTED WORKS•CASE 01•ai

All-in-One AI Workspace

Self-hosted LibreChat AI workspace orchestrated via LiteLLM proxy, connecting Google Vertex AI (Gemini 1.5 to 3.8 Flash, Nano Banana Pro), OpenRouter, and Replicate with mongo-data and meili-data persistence

ENTERPRISE AI WORKSPACELITELLM MULTI-ROUTER
ai.internal // workspace.stage
Unified AI Chat, Vision & Media Interface
Multi-provider routing across Vertex AI, OpenRouter, and Replicate with persistent mongo-data and meili-data search.● LIVE
✦ Gateway: LiteLLM proxy routing Vertex AI + OpenRouter + ReplicateHEALTHY
SYSTEM // LIBRECHAT

Multi-Model AI Gateway Hub

30+ frontier LLMs, vision engines, and private vector search under one zero-leak cockpit.

Unified AI Chat, Vision & Media Interface

Multi-provider routing across Vertex AI, OpenRouter, and Replicate with persistent mongo-data and meili-data search.

01 // PROVEN IMPACTVERIFIED
0+ Models

Frontier LLM and generative vision models unified under a single self-hosted ecosystem with zero vendor lock-in.

GATEWAY: LITELLM PROXY● 24/7 PRODUCTION
02 // THE CONTEXTPROBLEM ➔ FIX
THE CHALLENGE

Single-provider AI tools fragmented user history, lacked multimodal image/video generation, and suffered sluggish search.

THE SOLUTION

Engineered LiteLLM proxy bridge linking LibreChat to Vertex AI, OpenRouter, and Replicate with dual Mongo & Meili storage.

ZERO LONG PARAGRAPHS
WORKFLOW // LITELLM GATEWAY ROUTING
LIBRECHATXavierDevPrompt: Code & RAGLIBRECHATSarahDesignPrompt: 4K Video GenLIBRECHATMarcusCopyPrompt: VC SpeechLIBRECHATElenaDataPrompt: Metrics BILITELLMROUTER HUBPORT :4000ZERO-LEAKDISPATCH OKP1P2P3OPENROUTERFRONTIER LLM POOLGOOGLE VERTEXPRIVATE GATEWAYREPLICATEVISION & VIDEO APICLAUDE 3.7MODELCHATGPT 4oMODELGLM-4MODELDEEPSEEK R1MODELGEMINI FLASHMODELSEEDANCEVIDEONANO BANANAIMAGEGPT IMAGEIMAGEChatType a prompt…Shared workspace
ROUTING LATENCY: 38ms · ZERO-LEAK30+ MODELS ARMED ✓
02 // XAVIER'S ROLE & HOW HE BUILT ITAI GATEWAY & PLATFORM ARCHITECT

CASE STUDY ROLE: AI INFRASTRUCTURE ARCHITECT & LITELLM GATEWAY INTEGRATOR

Xavier took the open-source LibreChat platform and turned it into an enterprise-grade AI hub.

He introduced LiteLLM as the central smart router to seamlessly connect Google Vertex AI, OpenRouter, and Replicate, then configured a dual-database layer with MongoDB for chat persistence and Meilisearch for instant search.

WHAT XAVIER DIRECTLY DELIVERED & SHIPPED
Cloned open-source LibreChat and built a centralized LiteLLM proxy router to handle all incoming prompts
Connected Google Vertex AI for Gemini (1.5 to 3.8 Flash) and Nano Banana Pro models
Plugged in OpenRouter to give the team instant access to ChatGPT, Claude, GLM, DeepSeek, and MiniMax
Integrated Replicate API so any model can generate high-res images and videos directly within conversations
Configured mongo-data to securely store conversation threads and meili-data to index full-text search in < 4ms
HANDS-ON OWNERSHIPDIRECT ARCHITECT & BUILDER
03 // LANGUAGES & TOOLS USED
LANGUAGES & CORE RUNTIME
TypeScriptJavaScriptPythonYAML
FRAMEWORKS & ENGINES
LibreChat CoreLiteLLM ProxyTailwind CSS
TOOLS, AI MODELS & APIS
Google Vertex AI (Gemini 1.5–3.8 Flash, Nano)OpenRouter (ChatGPT, Claude, DeepSeek, MiniMax)Replicate (Image & Video Gen)
STORAGE & INFRASTRUCTURE
MongoDB (mongo-data)Meilisearch (meili-data)Docker Compose
CASE STUDY TECH ECOSYSTEM
LibreChatLiteLLM ProxyGoogle Vertex AIOpenRouterReplicateMeilisearch (meili-data)MongoDB (mongo-data)Docker Compose
TECH ARCHITECTURE● FULLY FUNCTIONAL
04 // HOW THE SYSTEM WORKS

HOW THE WORKSPACE OPERATES

4-STEP PIPELINE
STEP 01
01/04

Team Member Prompts

User types a question, uploads an image, or requests creative generation in the web chat.

StatusActive
STEP 02
02/04

LiteLLM Smart Router

Selects the optimal model (Gemini for speed, Claude for writing, DeepSeek for code, Replicate for visuals).

StatusActive
STEP 03
03/04

Multi-Cloud Generation

Securely queries Vertex AI, OpenRouter, or Replicate without exposing enterprise API keys.

StatusActive
STEP 04
04/04

Fast Answer & Dual Storage

Answers stream to the screen while mongo-data saves the chat and meili-data indexes it for instant search.

StatusActive