GenAI with TypeScript

A learning project exploring modern AI application development using TypeScript, focusing on state-of-the-art SDKs, frameworks, and architectural patterns.

Overview

The goal is to create examples needed to build advanced AI applications, including model routing, tool calling, RAG (Retrieval Augmented Generation), vector databases, evaluation frameworks, and agent-based systems.

Technologies Being Explored

• Vercel AI SDK – API and streaming tools for building AI apps.
• Mastra – Unified orchestration for AI agents and tools.
• Assistant UI – Pre-built UI components for building chat interfaces.
• Langgraph – Graph-based agent orchestration for complex workflows.

Core Concepts

• Model Routing
  Direct prompts or tasks to the appropriate model based on criteria like task type or expected output.

• Tool/Function Calling
  Enable AI models to call external functions/tools dynamically during interactions.

• Retrieval Augmented Generation (RAG)
  Enhance model outputs with external knowledge fetched at runtime, usually from vector databases.

• Vector Databases
  Store and retrieve embeddings to support semantic search and context injection for AI models.

• Evaluations (Evals)
  Techniques for systematically evaluating the performance and outputs of AI systems.

• Agents
  Systems where AI models can make decisions, call tools, plan actions, and reason over multi-step tasks.

Examples

• Text Completion
  Text completions using AI models.

• Text Completion (Streaming)
  Text completions using AI models with streaming responses.

• Image Analysis
  Attach images to get textual descriptions and analyses.

• Invoice Data Extraction
  Generate structured data from PDF files using AI models.

• Invoice Data Extraction (Streaming)
  Generate structured data from PDF using AI models with streaming response.

• AI Chat Assistant
  Interactive chat interface with streaming AI responses, perfect for building conversational AI applications.

• Weather Assistant
  Get real-time weather information using Tool Calls.

• Commit Message Generator
  Convert natural language descriptions into conventional commit messages.

Technologies Used

• React 19
• React Router
• TypeScript
• Tailwind CSS