How codebucks27/deep-research-ai-agent works
Project Type and Purpose
• This is a full-stack AI-driven research assistant built with Next.js (App Router) and the Vercel AI SDK.
• Purpose: guide users through an iterative “deep research” workflow—generating follow-up questions, crafting search queries, invoking LLMs and a Web Search API, then compiling a final research report.
File Structure (High-Level)
• src/app → Next.js UI routes and layout
• src/app/api → Serverless API endpoints (deep-research, generate-questions)
• src/components → Reusable UI components (Shadcn + custom)
• src/store → Client-side state (Zustand)
• src/lib → Utility functions
• Configuration → next.config.ts, tsconfig.json, .env.example
Main Workflow (Focus on Core Flow)
User → Next.js Page UI → “Start Research” trigger
↓
Client fetch to /api/deep-research (route.ts)
↓
API Handler calls service layer:
• activity-tracker.ts (logs progress)
• research-functions.ts (defines iterative loops)
• model-caller.ts (wraps Vercel AI SDK / OpenRouter to LLMs)
• constants/prompts/types (configure prompt templates)
• services.ts (orchestrates Exa Search API calls & LLM calls)
↓
External Services:
• LLMs (Gemini, GPT-4o via Vercel AI SDK/OpenRouter)
• Exa Search API
↓
API Response → Client UI
• Updates Zustand store (src/store/deepResearch.ts)
• Renders components (QnA, ResearchActivities, ResearchReport)