DEV Community: Debadatta Panda

Building an AI-Powered Product Chat with Amazon S3 Vectors and Nova Lite

Debadatta Panda — Thu, 08 Jan 2026 15:42:03 +0000

Vector search is quickly becoming the foundation of modern AI applications. Whether it’s product recommendations, semantic search, or conversational assistants, vectors enable systems to understand meaning rather than just keywords.

But building vector-powered systems hasn’t been easy:

Managed vector databases can cost thousands per month
Self-hosting open-source solutions means managing clusters, scaling, and performance
Scaling to millions of vectors introduces serious operational complexity

With Amazon S3 Vectors, AWS changes this equation by bringing vector search directly into Amazon S3 — no clusters, no capacity planning, no new infrastructure.

In this post, I’ll walk through how I built a serverless AI-powered product chat application using Amazon S3 Vectors, Amazon Bedrock (Nova Lite + Titan Embed), and AWS Lambda, following a clean RAG (Retrieval Augmented Generation) architecture.

What Is Amazon S3 Vectors?

Amazon S3 Vectors allows you to:

Store vector embeddings natively in S3
Perform high-performance similarity search (cosine similarity)
Scale automatically without managing infrastructure

GitHub Repository

Source code for the full implementation is available here:

https://github.com/debadatta30/awss3vector-productchat.git

Architecture Overview

This application is fully serverless and consists of:

Frontend: React app hosted on S3 static website
Backend: AWS Lambda
Vector Store: Amazon S3 Vectors
Embeddings: Titan Embed (Amazon Bedrock)
LLM: Nova Lite (Amazon Bedrock)
API Layer: Amazon API Gateway

Amazon S3 Vectors fundamentally simplifies how we build vector-based AI systems. When combined with Amazon Bedrock and serverless AWS services, you can build production-ready RAG applications that are:

Scalable
Cost-effective
Easy to operate
Free from infrastructure complexity

🎥 Video Walkthrough

https://youtu.be/yLfnRqcwEWo

Deploy Your Own AI Video Ad Generator on AWS (Using Nova / Poly / ECS / Streamlit)

Debadatta Panda — Wed, 03 Sep 2025 15:10:29 +0000

AI video advertisement creator that transforms text descriptions into complete video ads with voiceovers. Built with Streamlit + AWS Bedrock Nova + Polly. Features automated content strategy, image generation, video creation, and voice synthesis.

Building on my previous Streamlit prototype, this enhanced version now includes deployable-ready infrastructure:

✅ Full video pipeline (Bedrock Nova + Polly)

✅ Deployment with AWS CDK

✅ Container orchestration on ECS/Fargate

✅ CloudFront distribution + Cognito auth

There is a blog you can refer for more details https://debadatta30.medium.com/agentic-ai-with-strands-to-create-video-ads-from-your-prompts-af5f16395125

🏗️ Architecture Overview

The below diagram shows the complete architecture :

The application consists of:

Streamlit web interface with Strands agent integration
ECS/Fargate container deployment
Application Load Balancer with CloudFront distribution
Cognito user pool for authentication
S3 bucket for media storage
Integration with multiple AWS AI services

The complete source code and deployment instructions are available:
Repository: https://github.com/debadatta30/aws-ai-video-orchestrator
Quick start:

Clone the repo
Follow the detailed README

git clone https://github.com/debadatta30/aws-ai-video-orchestrator
cd aws-ai-video-orchestrator
cdk bootstrap
cdk deploy --parameters S3BucketName=your-bucket-name

Building an AI-Powered Video Ad Creator with AWS Nova and Strands Agents

Debadatta Panda — Tue, 19 Aug 2025 14:10:51 +0000

"Here's how I built a complete video ad creator using AWS's Nova models and Strands Agents: a 5-step AI pipeline that takes text input and outputs professional video with synchronized voiceover.

This is developed with the Strands Agents - an open-source SDK designed to make it dramatically easier to build such smart, autonomous systems

Creating a video ad involves multiple AI services that need to work together seamlessly. Here's how pipeline is designed with Strands Agent

Phase 1: Content Planning

# Input: "Luxury electric car driving through mountains"
# Output: Structured strategy for all subsequent steps

strategy = {
    "image_prompt": "Professional commercial photograph of luxury electric car on mountain road, golden hour lighting, cinematic composition, 1280x720",
    "video_prompt": "6-second commercial showing sleek electric car driving through scenic mountain curves, smooth camera tracking, sunset lighting, premium feel",  
    "audio_script": "Experience the future of driving. Luxury meets sustainability."
}

Phase 2: Visual Foundation

Service: Amazon Nova Canvas
Purpose: Create high-quality reference image that sets visual style
Output: Image stored in S3

Phase 3: Video Generation

Service: Amazon Nova Reel
Input: Video prompt + reference image
Process: Async generation (2-5 minutes)
Output: 6-second professional video footage

Phase 4: Voice Enhancement

Service: Amazon Polly Neural voices
Input: audio script
Output: Professional voiceover with natural intonation

Phase 5: Final Assembly

Tool: MoviePy + FFmpeg
Process: Merge video and audio with proper timing
Output: Complete video advertisement

Why This Tech Stack?

Strands Agents: AWS's new framework for building AI agents with a model-first approach
Amazon Nova: State-of-the-art multimodal models (Canvas for images, Reel for videos)
Streamlit: Rapid prototyping with beautiful, interactive UIs
S3: Reliable storage for all generated media files
Amazon Polly: Neural text-to-speech for professional voiceovers

You can refer the code in the github for this :

`# Clone the repository
git clone https://github.com/debadatta30/aws-strand-streamlit
cd aws-strand-streamlit

# Install dependencies
pip install -r requirements.txt

# Configure AWS credentials
aws configure

# Set up environment variables
cp .env.example .env
# Edit .env with your S3 bucket name

# Launch the app
streamlit run streamlit_agent.py`

AWS Permissions Required:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "bedrock:InvokeModel",
        "bedrock:StartAsyncInvoke",
        "bedrock:GetAsyncInvoke"
      ],
      "Resource": "*"
    },
    {
      "Effect": "Allow",
      "Action": ["s3:*"],
      "Resource": [
        "arn:aws:s3:::your-bucket-name",
        "arn:aws:s3:::your-bucket-name/*"
      ]
    },
    {
      "Effect": "Allow",
      "Action": ["polly:SynthesizeSpeech"],
      "Resource": "*"
    }
  ]
}

Bedrock Model Access:
Request access to these models in the AWS Bedrock console:

amazon.nova-canvas-v1:0 (Image generation)
amazon.nova-reel-v1:0 (Video generation)
us.amazon.nova-lite-v1:0 (Content strategy)