Building AI Apps: Data, Observability & Deployment (December 2025)

In Part 1, we covered SDKs and frameworks. Now let's complete the stack: data, observability, and deployment.

Layer 3: Data & RAG

When your AI needs to work with your data, you need a data layer.

LlamaIndex.TS

The RAG framework for TypeScript. Abstracts the complexity of loading, chunking, indexing, and retrieval.

2025 highlights:

LlamaParse with skew detection for better PDF extraction
LlamaSheets (Nov 2025) for parsing Excel and CSV files
Multi-document agent system for reasoning across collections
Azure AI Search integration with hybrid search
Runs on Node.js, Deno, Bun, and Cloudflare Workers

import { VectorStoreIndex, SimpleDirectoryReader } from 'llamaindex'
 
const documents = await new SimpleDirectoryReader().loadData('./docs')
const index = await VectorStoreIndex.fromDocuments(documents)
const queryEngine = index.asQueryEngine()
 
const response = await queryEngine.query('What is the refund policy?')

import { VectorStoreIndex, SimpleDirectoryReader } from 'llamaindex'
 
const documents = await new SimpleDirectoryReader().loadData('./docs')
const index = await VectorStoreIndex.fromDocuments(documents)
const queryEngine = index.asQueryEngine()
 
const response = await queryEngine.query('What is the refund policy?')

Best for: Document Q&A, knowledge bases, any RAG application.

Vector Databases

You need somewhere to store embeddings.

Pinecone

Fully managed, serverless, no ops required
Scales to billions of vectors
Official Node.js/TypeScript SDK
Best for: Production RAG at scale

Supabase pgvector

PostgreSQL with vector extension
Store relational and vector data together
Hybrid search (semantic + keyword)
Best for: Teams on Supabase, cost-conscious, moderate scale

Weaviate

Open-source with cloud option
Best-in-class hybrid search (vector + BM25)
Official TypeScript client
Best for: Complex search requirements

Quick decision:

Starting out → Supabase pgvector
Production scale → Pinecone
Hybrid search → Weaviate

Layer 4: MCP — The Industry Standard

MCP (Model Context Protocol) is the biggest shift in AI tooling in 2025. Created by Anthropic, it's now the industry standard.

What is MCP?

An open protocol that standardizes how AI models connect to external tools and data. Think of it as a universal adapter for AI integrations.

Why It Matters

Before MCP, every provider had different tool patterns. Now you write integrations once and they work everywhere.

2025 Enterprise Adoptions:

Atlassian Rovo — Jira and Confluence in any AI
S&P Global + AWS — Financial data for AI agents
Databricks Agent Bricks — Enterprise workflow automation
AWS API Gateway MCP Proxy — Convert REST APIs to MCP

Building with MCP

import { MCPServer } from '@modelcontextprotocol/server'
 
const server = new MCPServer({
  name: 'my-tools',
  tools: [
    {
      name: 'get_user',
      description: 'Get user by ID',
      inputSchema: { 
        type: 'object', 
        properties: { id: { type: 'string' } },
        required: ['id']
      },
      handler: async ({ id }) => {
        const user = await db.users.findById(id)
        return { content: JSON.stringify(user) }
      }
    }
  ]
})

import { MCPServer } from '@modelcontextprotocol/server'
 
const server = new MCPServer({
  name: 'my-tools',
  tools: [
    {
      name: 'get_user',
      description: 'Get user by ID',
      inputSchema: { 
        type: 'object', 
        properties: { id: { type: 'string' } },
        required: ['id']
      },
      handler: async ({ id }) => {
        const user = await db.users.findById(id)
        return { content: JSON.stringify(user) }
      }
    }
  ]
})

Use MCP from day one on any serious project. It future-proofs your integrations.

const openai = new OpenAI({
  baseURL: 'https://oai.helicone.ai/v1',
  defaultHeaders: {
    'Helicone-Auth': `Bearer ${process.env.HELICONE_API_KEY}`
  }
})

const openai = new OpenAI({
  baseURL: 'https://oai.helicone.ai/v1',
  defaultHeaders: {
    'Helicone-Auth': `Bearer ${process.env.HELICONE_API_KEY}`
  }
})