Skip to content
FAHD MOHAMED
All work

Professional work · RAG assistant · Custom CMS · 3D web

Agency Website + AI CMS

A 3D, scroll-animated agency website with a custom CMS and an AI assistant that only answers from published content.

chat providers
4
chat providers
source: src/server/ai/providers/*
dimension vectors
1536
dimension vectors
source: drizzle/0004_ai_assistant.sql
AI assistant admin: conversations, leads, answer rate, content gaps

The problem

Marketing teams need to edit everything themselves, and visitors increasingly ask questions instead of browsing. A generic chatbot would make up services and prices.

My role

  • Engineered the system: full-stack development, custom CMS, RAG assistant and AI integration

Engineered during my role as AI & Software Engineer at Elenor Marketing Agency, Cairo (Jan 2026 – present). Screenshots come from sanitized demo environments: client and business data are replaced with demo data.

What it is

A Next.js site with React Three Fiber, GSAP and Lenis, structured for search and AI engines (JSON-LD, sitemap, llms.txt). A custom CMS covers pages, services, case studies, a scheduled blog, FAQ, media, navigation, theme and users. The AI assistant retrieves from published CMS content only (pgvector embeddings, re-indexed on publish), streams answers, and captures and scores leads.

Key features

  • Retrieval-augmented assistant grounded only in published content
  • Provider switch: Claude, OpenAI, OpenRouter, Gemini; embeddings OpenAI or Gemini
  • Knowledge-health dashboard: indexed sources, gaps, top questions
  • Lead capture and scoring from chat
  • Full CMS with roles, scheduling (Vercel cron), drag-and-drop navigation, Tiptap editor

Architecture

Visitors

  • 3D site
  • Chat widget

Next.js

  • Public pages
  • Admin CMS
  • /api/ai/chat

AI

  • Retriever
  • LLM provider
  • Lead scorer

Data

  • Postgres + pgvector
  • Blob storage
  • Cron publish

AIRAG over CMS content with pgvector (real[] fallback)

AIStreaming answers

AILead extraction and scoring

Stack

Web

Next.js 14React Three FiberGSAPLenisTailwind

CMS

DrizzlePostgreSQL (Neon)TiptapVercel Blobdnd-kit

AI

ClaudeOpenAIGeminipgvector

Engineering challenges

01

Grounding by construction

The assistant can only see chunks indexed from published CMS records. When an answer is missing, it appears as a content gap in the admin dashboard instead of a guess.

Run locally against an anonymized copy of the seed content. No AI keys were connected, so the assistant admin shows its “not configured” state.

Knowledge base health and indexing
CMS dashboard · Client names anonymized
Public services page with animated 3D icons
Drag-to-reorder services
Navigation editor

Next project

Operations OS