I Built an AI-Powered Stock Analysis Platform with Next.js and Claude AI

After months of manually analyzing stocks and switching between multiple tools, I decided to build my own platform. Meet BistBase — an AI-powered stock analysis platform for the Turkish stock market (BIST).

🤔 The Problem

As a developer who trades on BIST, I was frustrated with:

• Switching between 5+ tabs for different analysis tools
• No AI-powered insights in Turkish market context
• Expensive premium tools with features I didn't need
• No backtesting capability for my strategies

💡 The Solution

I built BistBase — a full-stack platform that combines:

1. Real-time market data via Yahoo Finance API
2. AI analysis using Anthropic Claude SDK
3. Technical indicators (RSI, MACD, Bollinger Bands)
4. Backtesting engine for strategy validation
5. Portfolio tracking with email alerts

🛠 Tech Stack

Layer	Technology	Why?
Frontend	Next.js 16 + React 19	App Router, Server Components, great DX
Language	TypeScript 5	Type safety across full stack
Database	PostgreSQL + Prisma	Relational data, type-safe ORM
Cache	Upstash Redis	Rate limiting + data caching
AI	Anthropic Claude SDK	Best at understanding financial context
Auth	NextAuth 5	Flexible, built for Next.js
Charts	Lightweight Charts	TradingView-quality charts, lightweight
Monitoring	Sentry	Error tracking + performance
Styling	Tailwind CSS 4 + shadcn/ui	Rapid UI development

🤖 Why Claude for Financial Analysis?

I tested three AI models:

• Claude — Best at understanding Turkish market context. Long context window helps analyze larger datasets. ~85% accuracy on backtesting predictions.
• GPT-4 — Good general analysis but misses BIST-specific patterns.
• Gemini — Fast but less accurate for technical indicators.

The key insight: combine AI analysis with traditional technical indicators rather than relying solely on LLM output.

// Simplified AI analysis flow
const analysis = await anthropic.messages.create({
  model: 'claude-sonnet-4-20250514',
  messages: [{
    role: 'user',
    content: `Analyze ${stock.symbol} with these indicators:
      RSI: ${indicators.rsi}
      MACD: ${indicators.macd}
      Volume trend: ${indicators.volume}
      Provide buy/sell/hold recommendation with reasoning.`
  }]
})

📊 Architecture

┌─────────────────────────────────────────┐
│           Next.js 16 App Router         │
├──────────┬──────────┬──────────────────┤
│ Dashboard│ Analysis │ Backtesting      │
├──────────┴──────────┴──────────────────┤
│              API Routes (25+)           │
├──────────┬──────────┬──────────────────┤
│ Prisma   │ Claude   │ Yahoo Finance    │
│ (PgSQL)  │ SDK      │ API              │
├──────────┼──────────┼──────────────────┤
│ Upstash  │ NextAuth │ Sentry           │
│ Redis    │          │                  │
└──────────┴──────────┴──────────────────┘

🔑 Key Learnings

1. Rate limiting is crucial — Yahoo Finance API has strict limits. Redis-based rate limiter saved me from getting blocked.
2. Cache aggressively — Stock data doesn't change every millisecond. 30-second cache reduced API calls by 80%.
3. AI needs guardrails — Raw LLM output is unreliable for financial decisions. Always validate with real indicators.
4. Server Components = Performance — Moving data fetching to RSC reduced client-side JS by 60%.

📈 Results

• 25+ API endpoints
• Sub-200ms response times (thanks to Redis caching)
• ~85% accuracy on AI-assisted backtesting
• Dark mode because traders trade at night 🌙

🔗 Links

• GitHub: github.com/mvtandas/bistbase
• Author: Mustafa Vatandaş — Lead Frontend Developer

---

What AI-powered tools have you built? I'd love to hear about your experience with integrating LLMs into production apps. Drop a comment below!