BSO Space Blog

BSO Space Blog is a web-based Q&A system that answers questions based on blog articles using Retrieval-Augmented Generation (RAG).

Comment

Inspiration

Many Software Engineering students love learning through blogs yet often struggle to fully understand technical concepts or connect knowledge across topics.
We realized that while AI is often used to generate content, few systems focus on understanding content.
That inspired us to create BSO Blog, an intelligent blogging platform designed to make technical learning more interactive, accessible, and trustworthy.
Our vision is simple: “Be Simple but Outstanding.”

What it does

BSO Blog transforms a traditional blog into an AI-powered learning environment.
It offers two main Gen AI features:

1. BSO Bot Smart Blog Assistant

Users can chat directly with each blog post to summarize content, ask contextual questions, or explore related information.
The bot classifies user intent using AWS Bedrock’s reasoning model into four categories: Greeting/Farewell, Summarization, Question, and None of the Above.
For in-content questions, it runs a Retrieval-Augmented Generation (RAG) process within that post only using Amazon Titan Embeddings v2.
If the question is unrelated, it performs a web search and lists the top 10 relevant sites, blending precision and exploration.

2. AI-Powered Content Screening

Before publication, each post is reviewed by an AWS Bedrock reasoning model that detects harassment, hate speech, or harmful text.
Only approved posts can enable AI Mode, which triggers Titan Embeddings and unlocks the chat assistant.
Together, these features help students not only read but truly understand complex topics safely.

How we built it

Frontend: Next.js 15, Tiptap Editor, ShadCN UI, OAuth (Google/GitHub)
Backend: Go (Gin Framework), JWT Authentication, GORM, PostgreSQL
AI Stack:

  • AWS Bedrock meta.llama3-1-70b-instruct for reasoning, intent classification, and summarization
  • Amazon Titan Embeddings v2 for semantic search and RAG
  • Docker for CI/CD deployment

Data Flow:

  1. Author submits post → AI screening via Bedrock
  2. If approved → author enables AI Mode → Titan Embeddings generated
  3. User chats → Bedrock classifies intent → RAG (if needed) → contextual answer

Challenges we ran into

  • Designing a clean RAG pipeline that stays inside a single blog post while keeping performance smooth.
  • Managing embedding updates efficiently when blog content changes.
  • Balancing AI moderation and author freedom to ensure fairness.
  • Optimizing response latency between Bedrock and frontend streaming UI.

Accomplishments that we're proud of

  • Successfully integrated AWS Bedrock and Titan Embeddings into a full production blog workflow.
  • Built a context-aware chatbot that genuinely helps users understand technical topics.
  • Implemented real-time AI screening for content safety, ensuring ethical AI usage.
  • Delivered a complete, dockerized, and deployable solution for the hackathon.

What we learned

  • How to combine reasoning models and embedding models effectively for real-world education use cases.
  • The importance of intent classification before generation to improve accuracy and context.
  • Best practices for building responsible AI features that keep communities safe and productive.
  • Working with AWS Bedrock gave us a deeper understanding of scalable AI architecture.

What's next for BSO Space Blog

  • Expand RAG to include cross-blog knowledge and multi-source retrieval.
  • Add multilingual support (English and Thai) for summaries and answers.
  • Integrate AI learning analytics to track comprehension and improvement over time.
  • Deploy to a larger community of students under the BSO Space brand helping the next generation of developers learn smarter with AI.

Built With

Share this project:

Updates