Skip to main content
Web ApplicationCompleted

LayeredMindAI: Enterprise AI Chat Platform with Turkish Market Data

A multi-provider AI chat platform that puts hundreds of language models, real Turkish market data, multi-organization RBAC, and a Flutter mobile app for iOS and Android into a single deployable workspace.

DeveloperMustafa Kürşad Başer
Started: Oct 2025Completed: May 2026
LayeredMindAI: Enterprise AI Chat Platform with Turkish Market Data

LayeredMindAI is the AI chat platform an organization can actually run on. One workspace covers hundreds of language models across most major providers, structured Turkish market data through a custom MCP server, governed access for entire teams, and a Flutter mobile app that ships to both stores. It runs on the same infrastructure pattern an organization is already familiar with: a React web client, a Node.js backend, MongoDB plus Postgres, all packaged for everything from a single Docker host to a Kubernetes cluster.

The audience is concrete: asset management and capital markets research teams working in Turkish markets, enterprise teams that need governed multi-user access to a wide range of LLMs, research and support and engineering teams that lean on code execution and RAG and domain-specific tools, and product teams that want to ship their own branded AI assistants on web and mobile.

What the platform does:

  • Unified routing across OpenAI, Anthropic, Google Gemini, Azure OpenAI, AWS Bedrock, Vertex AI, Groq, Mistral, OpenRouter, Cohere, Deepseek, and local Ollama, with custom OpenAI-compatible endpoints for private models
  • Turkish market data through a dedicated Borsa-MCP integration, covering BIST stocks, Turkish investment funds, currency pairs sourced from TCMB, crypto, commodities, and a multi-country economic calendar, with technical indicators (RSI, MACD, Bollinger Bands, Pivot Points) surfaced as data rather than advice
  • Sandboxed code execution across Python, Node.js, Go, C/C++, Java, PHP, Rust, and Fortran, with file uploads, downloads, and per-session persistence
  • pgvector-backed RAG with document ingestion and file-aware context, so the assistant reasons over private knowledge bases instead of guessing
  • A custom sliding-window context optimizer designed to reduce token usage and downstream API spend on long conversations
  • A no-code agent builder with a marketplace, and Model Context Protocol integrations for teams shipping specialized assistants

Enterprise and security:

  • Multi-organization with tenant-isolated workspaces, fine-grained permissions on agents, prompts, and tools, and centralized usage tracking and billing
  • OAuth2 (Google, GitHub, Discord, Facebook, Apple), LDAP and Active Directory, SAML SSO, JWT email and password, and TOTP-based two-factor authentication
  • Mobile in-app purchases through RevenueCat, with the web tier handling tiered pricing, balance top-ups, and per-user, per-organization usage statistics directly against the backend
  • Conversation governance: branching chats, message forking, edit and resubmit, encrypted exports, and temporary privacy-mode sessions

Reach beyond the web:

  • A Flutter mobile app for iOS and Android with feature parity to the web, streaming chat, secure token storage, and offline-aware navigation
  • Live React and HTML artifacts with real-time preview, Mermaid diagrams, and image generation through DALL-E, Stable Diffusion, and Flux
  • Whisper and Azure speech-to-text plus OpenAI TTS and ElevenLabs voices for hands-free conversations
  • Wide multilingual coverage including full Turkish localization, RTL support, and dynamic UI translation through i18next

Stack and operations:

  • React on Vite for the client, Express.js on Node for the API
  • MongoDB for primary state, PostgreSQL with pgvector for embeddings, and Meilisearch for full-text search
  • An npm workspace monorepo with shared packages between web and tooling
  • Mobile: Flutter with Riverpod, Dio, GoRouter, Freezed, and a hand-rolled SSE client tuned for cellular networks
  • Deployment from local Docker Compose all the way up to Kubernetes Helm charts, with Nginx fronting TLS in production
  • Versioned MongoDB migrations with rollback companions for every schema change
  • Official @modelcontextprotocol/sdk over SSE, the Borsa-MCP server for Turkish market data, and a pluggable interface for adding new tool servers

The trade-off is deliberate: instead of locking organizations into a single closed AI platform, LayeredMindAI gives them a brand-able workspace that owns its data, its routing, and its compliance posture.

Tech Stack

React 18Node.js 20Express.jsMongoDBPostgreSQL + pgvectorMeilisearchFlutterRiverpodTailwindCSSViteDockerKubernetes + HelmModel Context ProtocolOpenAIAnthropic ClaudeGoogle GeminiAWS Bedrock