AWS Builder Center
Club Member Portal: A Grounded AI Knowledge Assistant for AWS Student Builders

Club Member Portal: A Grounded AI Knowledge Assistant for AWS Student Builders

A member-focused AI knowledge portal that helps AWS Student Builders find trusted club information through conversational search, source citations, and a continuously updated administrator-managed knowledge base.

Introduction

Joining a student community often means searching through multiple documents, announcements, shared drives, and communication channels to find simple answers such as:
  • When is the next workshop?
  • Where is today's event?
  • How do I publish on AWS Builder Center?
  • Who should I contact for technical help?
  • What are the hackathon rules?
We built the Club Member Portal to solve this information-fragmentation problem.
The goal is to provide new Student Builders with one trusted, members-only platform where they can:
  • Access club information
  • Ask questions naturally
  • Receive answers grounded in approved club documents
  • See the source behind every answer
The project started with the 70% baseline challenge and was then extended with the event-day 30% requirements:
  • Administrator document management
  • Live knowledge synchronization
  • Updated document lists
  • Smoke testing
  • Evaluator-facing API



What We Built

The Club Member Portal provides a complete member experience:
  • Email and password sign-up and login
  • Forgot-password and reset flow
  • Members-only AI chat
  • Source citations with answers
  • Safe fallback when information cannot be found
  • Persistent per-user chat history
  • Administrator document management
  • Document upload and paste support
  • Document updates and versioning
  • Automatic knowledge-index refresh after publishing
  • Live member document-list updates
  • Event-day briefing support
  • Three-question smoke-test validation
  •  POST /ask  API for external evaluation
The assistant is designed to answer only from the approved club knowledge base.
When relevant information cannot be found, it does not guess. Instead, it directs the member to the appropriate campus AWS Student Builder contact.



The Problem

Student communities frequently have information distributed across different documents and communication channels. This becomes particularly difficult during events when information changes quickly.
For example, an event schedule might originally say that lunch is at 12:30 PM and judging is in one room. If those details change on the day of the event, members may continue relying on outdated information.
The 30% challenge was designed around exactly this scenario:
Administrators need to publish updates, and the chatbot must use the new information within approximately 60 seconds without requiring members to refresh the application.



Our Solution

The portal creates a centralized knowledge layer between club members and the information maintained by the club.

Member Flow

Member
↓
Question
↓
Knowledge Retrieval
↓
Grounded Answer
↓
Source Citation


Administrator Flow

Admin
↓
Upload / Paste Document
↓
Publish
↓
Re-index
↓
Updated Knowledge
↓
Chatbot

This means administrators can maintain the knowledge base without rebuilding or restarting the entire application.



Grounded AI and Source Citations

A major design principle of the project is trustworthy answers over unrestricted generation.
The assistant uses the club's approved documents as its knowledge source.
Every response provides source information so that members can understand where the answer came from.
When the required information is not available, the assistant follows the project's fallback behavior rather than inventing an answer.

Example Fallback

"I could not find that in the club documents. Please contact Shanmukha Sasi Sadineni, AWS Student Builder Group Leader, at  sadinenisasi@gmail.com  or 7396025334."
This follows the project's requirement to avoid hallucinating information and direct members to the campus AWS Student Builder contact when retrieval is insufficient.



Administrator Knowledge Management

The event-day extension adds an administrator workflow to the original member portal.
An authorized administrator can:
  1. Open Manage Documents
  1. Create or select a document
  1. Paste Markdown/text or upload a document
  1. Edit existing content
  1. Publish the document
  1. Trigger knowledge-index refresh
  1. See the updated document status
The important part is that publishing does not require restarting the server.
When a document is updated, the changed knowledge is incorporated into the in-memory retrieval index and becomes available to subsequent chatbot requests.



Efficient Knowledge Retrieval

For the local hackathon implementation, we use a lightweight in-memory lexical retrieval approach.
The approved knowledge sources are loaded into memory rather than being read from disk for every chatbot question.

Initial Flow

Application Startup
↓
Load Knowledge
↓
Build In-Memory Index
↓
Serve Questions


After Administrator Publishing

Publish
↓
Persist Document
↓
Update Index
↓
New Content Available

This avoids repeatedly reading the same source files for every question and keeps local retrieval fast and predictable during the hackathon demonstration.
For a production deployment, this retrieval layer can be replaced with an embedding-based vector retrieval architecture without changing the member-facing portal or the evaluator API.



Chat History

The portal also maintains per-user chat history.
Members can return to previous questions and answers instead of losing their conversation after leaving the page.
Chat history is associated with the authenticated member, allowing the application to maintain a personalized conversation experience while keeping authentication and authorization separate from the knowledge base.



Event-Day Knowledge Synchronization

The event-day document contains updated information such as:
  • Lunch at 1:30 PM
  • Judging check-in at 3:00 PM
  • Final demos at 4:00 PM
  • Judging and final demos in Hall B, Room 204
These values come from the event-day briefing document.

Intended Demonstration Flow

Administrator publishes an update
↓
Knowledge index refreshes
↓
Member document list updates
↓
Member asks a question
↓
Chatbot retrieves updated information
↓
Answer cites the document

This is the core knowledge-synchronization requirement of the 30% challenge.



External API

The portal exposes the required:
POST /ask

endpoint on port 8080.

Request

{
"question": "What time are final demos?"
}


Response

{
"answer": "Final demos are at 4:00 PM.",
"sources": [
{
"document": "event-day-briefing.md",
"rank": 1
}
]
}

This follows the required evaluator contract, where  answer  and  sources  must be top-level fields rather than being wrapped inside another object.



Technology Stack

Frontend

  • HTML
  • CSS
  • JavaScript

Backend

  • Node.js
  • HTTP API
  • REST-style endpoints

Local Data Storage

The hackathon prototype uses a lightweight JSON-based local data store for application data, including:
  • Users
  • Sessions
  • Chat history
  • Document metadata

Knowledge Retrieval

  • In-memory knowledge index
  • Lightweight lexical retrieval
  • Document-level source attribution
This approach keeps the local prototype simple, fast, and dependency-light.



AWS Production Architecture

Although the hackathon application can run completely locally, the project is designed with an AWS deployment path in mind.
A production architecture could map the current components to AWS services:
Current Requirement
AWS Service
Authentication and role management
Amazon Cognito
Password-reset emails
Amazon SES
Document storage
Amazon S3
Backend APIs
AWS Lambda / API Gateway
Foundation model and generation
Amazon Bedrock
Document and query embeddings
Amazon Titan Text Embeddings
Vector retrieval
Amazon OpenSearch Serverless / Amazon Bedrock Knowledge Bases
Users, metadata, chat history, application state
Amazon DynamoDB

Production RAG Flow

Amazon S3
↓
Document Ingestion
↓
Chunking
↓
Embeddings
↓
Vector Store
↓
Retrieval
↓
Amazon Bedrock
↓
Grounded Answer + Sources

This follows the AWS architecture direction described in the project reference, where S3, embeddings, OpenSearch Serverless or Bedrock Knowledge Bases, retrieval, and Bedrock generation form the production RAG workflow.



Security and Responsible AI

Security and responsible AI were important design considerations.
The portal includes:
  • Password hashing
  • Session-based authentication
  • Members-only chat
  • Server-side administrator authorization
  • Separation between member and administrator functionality
  • No intentional logging of passwords or private student information
  • Document-grounded responses
  • Source citations
  • Fallback behavior instead of unsupported claims
For a production deployment, authentication and role management would be strengthened further through Amazon Cognito and managed AWS identity controls.



Smoke Testing

The event-day implementation includes three smoke-test questions designed to verify that the chatbot correctly retrieves information from the updated event-day document.
The smoke test validates:
  • Correct answer
  • Correct source document
  • Successful retrieval after document updates
This provides a simple way to demonstrate that knowledge synchronization is working rather than relying only on the UI.



Why This Project Matters

The Club Member Portal is more than a chatbot.
It demonstrates how an AI assistant can be connected to a controlled organizational knowledge base while giving administrators a practical way to keep that knowledge current.
The project addresses three important problems:

1. Information Fragmentation

Members no longer need to search through multiple documents and communication channels.

2. Information Freshness

Administrators can publish updated information without restarting the application.

3. AI Trust

Answers are grounded in approved documents and accompanied by source citations, while unsupported questions trigger a safe fallback.



Future Improvements

The current prototype provides the required foundation, but the architecture can be extended further.
Future versions could include:
  • Multiple administrator roles using Amazon Cognito groups
  • Full audit history for document changes
  • Document processing status such as  Uploaded → Processing → Indexed 
  • Semantic chunking
  • Embedding-based retrieval
  • OpenSearch Serverless
  • Amazon Bedrock Knowledge Bases
  • S3-based document storage
  • DynamoDB for scalable application data
  • CloudWatch monitoring and observability
  • CI/CD deployment
  • Production AWS deployment
  • Advanced analytics for frequently asked questions



Conclusion

The Club Member Portal demonstrates how a student community can move from static documents to a continuously maintainable AI-powered knowledge system.
Members get a simple conversational interface.
Administrators get control over the knowledge base.
The system provides source citations instead of opaque answers, and newly published information can become available without taking the application offline.
Our local implementation focuses on reliability and transparency for the hackathon, while the architecture provides a clear path toward a production AWS solution using:
  • Amazon Cognito
  • Amazon S3
  • Amazon Bedrock
  • Amazon OpenSearch Serverless
  • AWS Lambda
  • Amazon API Gateway
  • Amazon DynamoDB
  • Amazon SES
Ultimately, our goal is simple:
Make it easier for every new Student Builder to find the right information, trust the answer, and know where to go when the answer is not available.
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article