IT Square

Data flow and AWS deployment architecture
Kiro to build requirement, design, implementations
High level program design diagram
Project prototype website: it-square.hk

Inspiration

Hong Kong’s media industry faces a unique set of challenges:

Fragmented, Multilingual Sources

Editors receive press releases and event invites in Traditional Chinese, Simplified Chinese, and English—often as unstructured emails or images.

Limited Resources vs. High Volume

Small editorial teams struggle to localize, fact-check, classify, and publish quickly in a fast-moving tech ecosystem.

Legacy CMS Overhead

Systems like WordPress create security risks, require heavy maintenance, and slow down content delivery.

Demand for Real-Time Relevance

Readers expect up-to-date, mobile-first content and event listings, but current workflows can’t keep up.

Our Inspiration

We asked ourselves: “What if AI could handle the heavy lifting—so editors can focus on storytelling?” That question inspired us to build an AI-assisted content editing and publishing platform, leveraging AWS infrastructure for speed, security, and scalability.

What it does

Our platform uses AI and AWS services to transform raw, multilingual content into validated, personalized news—delivered instantly across web and mobile.

How we built it

Utilize AWS Kiro and Q Developer for Python coding. Perplexity LLM for AI interface. AWS infrastructure for serverless components and Next.js integration for content rendering.

Generated production grade code for AWS serverless components and Next.js integrations, reducing development time and enabling our small team to build enterprise-grade features that would be impossible without AI.
Built an RPA like, multimodal extraction workflow: parses free form invitation emails and images, normalizes fields, verifies addresses with open data, and outputs canonical JSON for DynamoDB. This system handles Hong Kong's unique addressing system with building names, floors, and units, using spatial data from Hong Kong SAR Government.
Integrated via MCP to interact with AWS services and Bedrock KB, automating all configurations and service settings, including API and database connections, without the headache of IAM, role, and permission settings. This was critical for our rapid development timeline

Challenges we ran into

Time-Critical Development

With limited time, defining a clear MVP scope is essential to deliver a functional prototype quickly.

Infrastructure Design

Building a flexible and scalable architecture is critical to ensure performance and future growth.

AI/LLM Evaluation

We need to test multiple AI/LLM models to verify that their logic and outputs meet editorial and compliance requirements.

Frontend Framework Selection

Choosing the right framework is key to achieving the desired UI/UX. After evaluation, we selected Next.js for its performance and developer experience.

Accomplishments that we're proud of

Despite a very short timeline, we successfully delivered a working prototype while simultaneously learning and adapting to a completely new development toolset and environment.

Why Our Solution Fits Hong Kong

Automates bilingual normalization and metadata so small teams can cover more HK specific news and events, with localization in style and news angles. Our system understands Hong Kong-specific terminology and adjusts tone for local audiences.

Extracts structured event data from messy media invites common in local PR workflows. The system automatically recognizes Hong Kong venues, districts, and transportation options. Serverless architecture lowers cost and attack surface, ideal for lean local publishers facing Hong Kong's high operational costs and competitive digital media landscape.

Personalization elevates relevant local content and events to readers, recognizing Hong Kong's diverse tech interests from fintech and Web3 to hardware manufacturing and mainland China tech giants.

What we learned

Rapid Prototyping Under Time Constraints

We learned how to define a clear MVP scope and prioritize features to deliver a working prototype within a very short timeline.

Building on AWS Infrastructure

Gained hands-on experience with AWS services such as S3, Lambda, AppSync, DynamoDB, and CloudFront, and learned how to integrate them into a scalable, secure architecture.

Leveraging AI/LLM for Content Processing

Explored multiple AI/LLM models to sanitize and structure multilingual content, and understood the importance of validating logic and compliance with editorial rules.

Adopting AWS Kiro for Development Acceleration

Learned how AWS Kiro streamlines development through spec-driven workflows, automated documentation, and enterprise-grade governance.

Frontend Framework Selection and Optimization

Evaluated different frameworks and finalized Next.js for its performance and developer experience, ensuring a responsive, mobile-first UI.

Collaboration and Adaptability

Improved our ability to adapt to new tools and environments quickly, while maintaining team alignment and delivering results.

What's next for NewsFlow AI

Enhanced AI Accuracy

Fine-tune LLM models for better multilingual handling and editorial compliance. Incorporate domain-specific training for media and event content.

Advanced Personalization

Expand Amazon Personalize integration for deeper behavioral insights and predictive recommendations.

Content Moderation & Compliance

Implement automated fact-checking and compliance validation using additional APIs and AI pipelines.

Scalability & Performance Optimization

Introduce serverless optimizations and caching strategies for high-traffic scenarios. Explore multi-region deployment for better latency and redundancy.