✨ Powered by Amazon Bedrock

GenAI Chat
Platform

A serverless AI chat application built on AWS. Converse with foundation models through a secure, production-grade interface.

🧠

6 Foundation Models

Claude, GPT-OSS, GLM — pick the right model for every task

⚙️

Prompt Engineering

Custom system prompts, few-shot examples, and reusable presets

📊

Usage Analytics

Track tokens, cost, and usage per user and model

🔐

Secure by Default

Cognito auth, CloudFront CDN, private S3 origin

⚡

Serverless Stack

Lambda + API Gateway — zero servers to manage

🤖

Welcome Back

Sign in to start chatting

Available Models
Haiku 4.5 GPT-OSS 20B Sonnet 4.6 GLM 5 Opus 4.6 Opus 4.7 Qwen 2.5 3B ⚡
CloudFront · Lambda · Bedrock · DynamoDB

🤖 GenAI Chat

System Prompt Sets the AI's persona, rules, and behavior
0 chars
Few-Shot Examples Teach the model by example (user → assistant pairs)
⚠️ Rate limit reached. Retrying in 5s — or switch to a different model.

Hello!

I'm powered by Amazon Bedrock. Send me a message and I'll respond using Claude.

💡 Tip: Click ⚙️ Prompt below to set a system prompt or add few-shot examples.

⚖️ Model Comparison

🏆 Rankings

📈 Performance Metrics

🧪 Simulation Progress

💬
—
Requests
⚡
—
Avg Latency
💰
—
Total Cost
🏆
—
Fastest Model
Model Performance Comparison
Model Requests Avg TTFT Avg Total Avg Tok/s P50 P95 Cost
Loading metrics...

Latency Trend

Request Volume

Application Architecture

🧑‍💻 User Browser 🛡️ WAF Web ACL Managed Rules ☁️ CloudFront CDN Distribution /* → S3 /api/* → API Gateway 🪣 S3 Bucket Static Frontend HTML / CSS / JS 🔐 Cognito User Pool OAuth2 / JWT 🔀 API Gateway REST API /api/chat · /api/usage Cognito Authorizer ⚡ Lambda Node.js 22 (ARM64) Chat + Usage API 🧠 Bedrock Converse API Claude Models 📊 DynamoDB Usage Tracking Per-user / Per-model 💰 Cost Explorer Billing Data Actual AWS Cost HTTPS Allow OAC /api/* Proxy Converse Track Query cost Validate JWT Login redirect

🛡️ AWS WAF

Web Application Firewall protecting the CloudFront distribution with AWS managed rule groups and rate limiting.

☁️ CloudFront

CDN distribution with behaviors: default serves static files from S3, /api/chat, /api/usage, and /api/metrics route to API Gateway. HTTPS enforced, HTTP/2 enabled. WAF-protected.

🪣 S3 Bucket

Private S3 bucket hosts the static SPA frontend. All public access blocked — served exclusively through CloudFront via Origin Access Control (OAC).

🔐 Cognito

Shared Cognito user pool with OAuth2 authorization code flow. Browser redirects to Cognito Hosted UI, exchanges code for JWT tokens. No client secret (public SPA client).

🔀 API Gateway

Regional REST API with Cognito authorizer validating JWT on every request. POST /api/chat for conversations, GET /api/usage for analytics, GET /api/metrics for performance data. CORS enabled.

⚡ Lambda

Node.js 22 on ARM64 (Graviton). Handles chat (Bedrock ConverseStream + InvokeModel for custom imports), usage tracking (DynamoDB writes), metrics queries, and usage queries (DynamoDB reads + Cost Explorer).

🧠 Amazon Bedrock

ConverseStream for native models (Claude, GPT-OSS, GLM) and InvokeModel for Custom Model Imports (Qwen 2.5 3B). 7 total models. Real-time TTFT measurement via streaming.

📊 DynamoDB

Two tables: genai-chat-usage for monthly aggregates and genai-chat-metrics for per-request performance data with 90-day TTL.

💰 Cost Explorer

Queried for actual AWS Bedrock billing data. Provides real cost alongside token-based estimates. Data has 24–48h delay from AWS.

📊 Monthly Usage

💬
—
Requests
📥
—
Input Tokens
📤
—
Output Tokens
💰
—
Estimated Cost
☁️
—
Actual Bedrock Cost
Breakdown by Model
ModelRequestsInput TokensOutput TokensEst. Cost
Loading usage data...
Breakdown by User
UserRequestsInput TokensOutput TokensEst. Cost
Switch to "All Users" view