Premium Mode

Chat with My AI Twin

How This Was Built

The design and delivery story behind the digital twin experience, from UI system to Terraform-backed deployment.

How This Was Built

Build Pipeline

01

Experience Design

Shared premium UI system, glass surfaces, consistent navigation, and branded gradients across every route.

02

3D Cosmic Engine

React Three Fiber scene with round stars, neural pulses, bloom, nebula layers, and motion tuned for cinematic depth.

03

AI Twin Interface

Client-side twin experience with voice input, Polly voice output, file uploads, scheduling flow, interactive clickable follow-up questions, and scroll-safe streaming UX powered by Claude 3.5 Sonnet v2.

04

Voice Layer (Amazon Polly)

Assistant responses are synthesized to lifelike speech with Amazon Polly (British male profile) and streamed back to the browser.

05

Video Privacy Pipeline

End-to-end video redaction: Python script runs Rekognition text detection on extracted keyframes (1-frame per 2s) to identify attendee names and IP patterns. FFmpeg complex filters apply selective boxblur (40px radius) to detected text regions only, preserving overall video clarity and host visual identity. H.264 re-encoding with faststart flag enables efficient CloudFront CDN streaming. Privacy-processed videos are stored in S3 and cached globally.

06

Infrastructure as Code

Terraform drives AWS resources, certificates, domains, storage, API edge routing, and deployment outputs.

07

CI/CD and Delivery

GitHub Actions builds, validates, deploys, and triggers CloudFront refresh for production delivery.

Build Sequence

Design system → 3D cosmic engine → AI twin interface → Claude 3.5 Sonnet v2 (Bedrock) + Polly integration → interactive follow-up questions → Terraform-defined backend → CI/CD automation → CloudFront production delivery.

Production Outcomes

Frontend

Next.js + React

3D Layer

R3F + Drei

Cloud Stack

Terraform + AWS

Frontend delivery

Browser

User visits the site over HTTPS

CloudFront

Global CDN, HTTPS termination, edge caching

S3 — Static Frontend

Next.js app, statically exported

Chat backend

API Gateway (HTTP API)

Routes /, /health, /chat, /tts, /contact, /schedule-meeting

Lambda — FastAPI

Python 3.12, Mangum adapter

Amazon Bedrock + Claude 3.5 Sonnet v2

anthropic.claude-3-5-sonnet-20241022-v2:0 — contextual multi-turn reasoning with session memory

Amazon Polly

British male voice synthesis for assistant speech output

DynamoDB + SES

Contact capture, scheduling requests, and confirmation emails

CloudWatch

Operational metrics and telemetry for chat and scheduling flows

S3 — Memory Bucket

Per-session conversation history (JSON)

Amazon Rekognition

Frame-level text detection to identify attendee name labels for privacy redaction in meeting recordings

S3 — Video Bucket

Rekognition-processed meeting recordings with attendee names blurred, served via CloudFront CDN

Infrastructure and delivery

Every resource above is defined as code in Terraform, with remote state in S3. Pushing to main triggers GitHub Actions that builds Lambda in a runtime-matched container, applies Terraform, rebuilds the frontend, syncs static output to S3, and invalidates CloudFront through short-lived OIDC credentials. The production assistant combines Claude 3.5 Sonnet v2 via Amazon Bedrock for reasoning, interactive follow-up question buttons, and Amazon Polly for natural British male speech output.

Complete App Capabilities

  • • AI chat powered by Claude 3.5 Sonnet v2 via Amazon Bedrock — contextual multi-turn reasoning with session memory
  • • Amazon Polly speech synthesis for assistant voice playback with British male accent
  • • Browser Web Speech API for voice input and hands-free conversational mode
  • • File upload analysis workflow with multi-modal evidence injection
  • • Meeting scheduling system with iCal integration and confirmation emails via SES
  • • Interactive follow-up questions rendered as clickable buttons after each response — no typing required
  • • Session memory continuity with DynamoDB persistence per-user
  • • Premium 3D cosmic UX with Three.js neural network visualization and bloom effects
  • • End-to-end video privacy: Rekognition text detection + FFmpeg selective blur + S3 + CloudFront
  • • HTTPS everywhere with ACM certificates and Route53 DNS management

Deployment Runbook

  • • Code is versioned in GitHub with environment-aware Terraform variables.
  • • CI/CD packages Lambda in a runtime-matched container and applies infrastructure.
  • • Frontend static export is published to object storage and delivered through CDN.
  • • Backend API is exposed through HTTP edge routing with strict CORS controls.
  • • Observability comes from CloudWatch metrics/logs and deployment telemetry.

This architecture pattern is cloud-portable: static frontend + CDN, serverless API runtime, managed model endpoint, object storage for memory, and managed telemetry.

Cloud Security

Zero-trust architecture.

Designing and implementing enterprise-grade security controls across cloud platforms—encryption, identity, compliance, and threat detection.

  • • AWS Security Specialty, Azure Expert
  • • HIPAA, SOC2, FedRAMP compliance
  • • Threat modeling & penetration testing
  • • WAF, DLP, GuardDuty, Security Hub
View services

DevOps & Infrastructure

Code-driven ops.

Infrastructure-as-Code, CI/CD pipelines, Kubernetes orchestration, and automated deployments—eliminating manual provisioning and human error.

  • • Terraform, CloudFormation, CDK
  • • GitHub Actions, CodePipeline, GitLab CI
  • • EKS, AKS, GKE administration
  • • 99.95%+ uptime SLA track record
View projects