WHAT SHOULD I HACK?

Put your skills to the test and apply Azure AI, Power Platform, Cybersecurity and Data Services to a solve one or more of the following problem statements. This is an individual hackathon and no teams are required. Projects may use Azure services, open source technologies (including but not limited to frameworks, libraries, and APIs) and physical hardware of your choice. 

To get access to the required tools, Makers must sign up for the Azure free trial or use an existing Azure. New accounts will automatically receive $200 in Azure credits to use towards building submission applications. If you previously created an Azure free trial account, please register again using a new email address.

Register for a fully sponsored AI102 training here - https://forms.microsoft.com/r/N75KfkP2v9 

 

Partner Problem Statements

BAE Systems

1. Current State: There are many research articles on cyber threat actors coming from different sources (independent researchers, security vendors, etc.). Oftentimes a single threat actor is described under different names (due to different naming conventions, etc.); as a result, there are often commonalities between reports despite the different names. Sometimes there are also relationships between different threat actors (e.g. A is a subgroup of B; A facilitates B).

Problem Statement: Generate a definitive threat actor database/dossier - sorting out all namings (aliases), relationships, attribution confidence, etc. Are there trends that can help guide further research and effective maintenance of this database/dossier.

Data Source: Public threat research blogs and articles

• Here are some websites well known which would routinely report on threat actors:

https://www.microsoft.com/en-us/security/blog/microsoft-security-intelligence/

https://www.crowdstrike.com/blog/category/threat-intel-research/

https://www.trendmicro.com/en_us/research.html?category=trend-micro-research:threats/apt-and-targeted-attacks

https://symantec-enterprise-blogs.security.com/blogs/threat-intelligence

Hint: Text analytics (entity extraction), Cosmos DB, knowledge graph

2. Current State: Researchers want to keep track of adversary techniques used by threat actors. There is a knowledge base/catalogue of techniques (MITRE ATT&CK), but the techniques are often discussed in prose form within threat research articles. Sometimes a list of techniques is provided at the end of the article, but these tend to be manually generated by the article’s authors.

Problem Statement: Given a threat research report/article, generate, using AI methods, a list of MITRE ATT&CK techniques described in the article. Make the techniques searchable.

Data Sources:

• https://attack.mitre.org/techniques/enterprise/ 
• Public threat research blogs and articles

Hint: Text Analytics, Cognitive Search

GITHUB

1. AI Chat bot that can teach you Git and GitHub. Specialized service trained on the open source git-scm book and open source GitHub Docs data sets along with a recent stackoverflow archive to help you learn how to contribute to an open source project and how to use all elements of GitHub.
Hint: Bot Framework

2. Markdown alt text suggestor - GitHub Action that can be enabled by a maintainers to scan markdown documents looking for inline images missing alt text and suggesting them. This can be run as part of a PR pre-merge check or periodically on the codebase to help improve the accessibility of a project documentation set
Hint: Computer Vision

HCLTech

Influenza outbreak event prediction Participants are expected to predict whether or not there is an influenza outbreak on the following week in each region using the Twitter Dataset or  Influenza Surveillance Dataset

Hint: Azure ML

Johnson & Johnson

1. Predicting Length of Stay in ICU using MIMIC-III Dataset

Problem Description: Intensive Care Units (ICUs) provide critical care to patients with life-threatening illnesses and injuries. Accurately predicting the length of stay (LOS) of patients in ICU can help healthcare providers allocate resources efficiently and improve patient outcomes. In this challenge, participants will use the publicly available MIMIC-III dataset to develop models that predict the LOS of ICU patients.

Dataset: MIMIC-III (Medical Information Mart for Intensive Care III) is a large, freely-available database comprising deidentified health-related data associated with over forty thousand patients who stayed in critical care units (https://mimic.mit.edu/docs/iii/) 

 Here are some of recommended steps for approaching the problem:

• Getting access to the dataset
• Understand the dataset including the tables, meaning of the attributes, etc.
• Conduct EDA (explanatory data analysis) to visualized the key indicators impacting ICU length of stay
• Build a predictive model to predict the length of stay in ICU
• Design methodology to verify your prediction and present the result.

Hint: Azure Machine Learning

2. Mall Analytics for Commercial Success

 Problem Description: It is crucial to understand visitor traffic, customer demographic attributes and market trends to increase commercial success. Please develop a model to identify popular malls that are likely receive high foot traffic from the public. Deductions about the age group and customer profile (such as affluence) who may visit these malls are highly valuable as well.  

 Here are some attributes you may consider, although other methods demonstrating innovation & creativity are welcome as well:

• Distance to Malls, MRTs, Residential areas
• Footfall traffic at nearby MRTs & bus stations
• Type of amenities in surrounding area, and distance to these amenities
• Number of parking lots at mall
• Google API data showing “popular timings” which has more visitors on an average day (e.g. https://github.com/m-wrzr/populartimes)

Other possible data sources you may use (but not limited to) includes:

• Data published by LTA on Public transport: https://datamall.lta.gov.sg/content/datamall/en/dynamic-data.html
• Data published by Gov on Carparks: https://data.gov.sg/dataset/carpark-availability 

Hint: Azure Machine Learning

 

PALO IT

1. Hospital throughput prediction

Context: Bottlenecks and increased waiting times at public health institutions is an ongoing issue in most communities.

• Some patients have to wait up to 50 hours for beds - Longer waiting times at hospitals with some patients told to wait up to 50 hours for a bed - CNA (channelnewsasia.com)
• The average waiting time for Emergency Departments (ED) varies from hospital to hospital but can be as high as 5 hours for non-critical cases https://www.moh.gov.sg/resources-statistics/healthcare-institution-statistics/waiting-time-for-admission-to-ward

Problem Statement:

• How might we predict the average waiting time of a hospital based on the patient’s medical history and severity of conditions and simultaneously allocate the most at risk patients to the hospital with optimal waiting times. IE: We want to find a way to dynamically allocate a patient to a hospital that we have predicted will have the lowest waiting time based on historical data according to the patient’s symptoms and previous medical history. Highest risk to be served first
• We will need to determine which hospitals are the most efficient
• We will need to know which patients should be prioritized
• We will need to train and test the model with additional simulated data based on the sample dataset

Datasets:

• Sample Admissions Dataset – https://www.kaggle.com/datasets/ashishsahani/hospital-admissions-data
• MOH Resources and Stats for mapping hospital admission times - https://www.moh.gov.sg/resources-statistics

Hint: Azure ML

2. Patient Disbursement to community healthcare network

Context:Bottlenecks at hospitals could potentially be reduced through preventative care

• As part of Healthier SG there is a drive to connect and strengthen the relationship between local physicians (GPs) and the community with the idea that consistent and continuous healthcare at community level may reduce the occurrence of more severe illnesses – https://www.moh.gov.sg/news-highlights/details/promoting-overall-healthier-living-while-targeting-specific-sub-populations

Problem Statement:

• How might we monitor at-home / community sourced vitals and use these insights to autonomously schedule GP visits and generate generic care plans. We can limit this to only diabetes patients at this stage
• We will need to create a simple health form for input parameters (Geography, Time Preference, Diet, Medication Adherence etc.)
• We will have to use a dataset of illness, diabetes in. this case, to predict onset or pre-diabetes
• We will use the information above to schedule appointments with Doctors that suit location and time preferences
• We will generate lifestyle and diet changes according to risk of onset diabetes

Datasets :

• Diabetes prediction - https://www.kaggle.com/code/ahmetcankaraolan/diabetes-prediction-using-machine-learning/data 

Hint: Azure ML, Power Apps

 

Petronas

There are 8 different types of publicly-available PETRONAS reports (i.e. Integrated & Annual Reports, Financial Reports & Sustainability) These reports contain a wealth of information, but their complex format and large volume make it challenging for users to quickly identify key topics and generate insights. How can we use Microsoft AI-related services to develop a solution that can automatically extract and organize relevant information from these PETRONAS reports to help users quickly find and understand the topics that they are interested in?

Success Criteria:

1. Leverage on Microsoft AI-related services to extract and categorize text and images from PETRONAS reports, and identify key topics within each report.
2. The scope for this hackathon is on Integrated & Annual Reports and Sustainability. However any additional reports that can be included in the solution will be considered a bonus.
3. Develop a landing page with search bar that utilizes Natural Language Understanding (NLU) to allow users to search for topics of interest within the reports
4. Upon a search query, the tool should surface relevant documents related to the query and highlight specific keywords from the content across multiple reports.
5. The tool should also generate a visual representation of relevant entities in a knowledge-graph with their relationships to help users better understand the context of the topics they are interested in.

Hint: Cognitive Services

PTP 

PTP is in need of a modern Fuel Management System (FMS) that can effectively keep track of fuel consumption for its equipment (Prime Mover, RTG, etc). The current system is manual and prone to errors, making it difficult for the company to accurately analyze the fuel consumption trends and costs involved.

To address this issue, PTP is seeking solutions from skilled developers and data scientists to create a new FMS that will be equipped with advanced features such as Power Apps and AI Vision. The goal of the hackathon is to:

1. Develop an online Power Apps platform for pump station operators to use, which will allow them to scan equipment IDs (on the vehicle), input operator IDs, scan fuel consumption values (on the pump station), and store the data in a database. The requirement for the new FMS is as below:

• The Power Apps platform should be equipped with AI Vision that can convert images of Equipment ID and Fuel consumption values captured into a text input. The sample image can be obtained here LINK.
• Developers must design the system in a way that the system allows for analytics on trend, consumption, and other potential related analysis
• The database used must be a SQL database. (Preferably MSSQL, but any SQL DB will do)
• Provide simple data visualizations (reports/dashboards) using the data obtained from the system to monitor fuel input-output.

PTP is eager to implement a modern Fuel Management System that will help them to make better business decisions and reduce its carbon footprint.

Hint: Cognitive Services, Power Apps, Power BI, SQL

General Problem Statements

Schools are getting more data driven, we have such a school that has been collecting data across years to understand results, enrollments across states and districts.

They would like to consolidate all the data they have across years and understand how their report cards or results have changed over the years. They understand the benefit of the cloud, and create a cloud based system to help them with their needs.

The challenge is to help build a scalable and cost optimized solution to help the school with the following:

1. Ingest multiple years data, ensure to select the storage format that helps with good compression of the data. The development team at the school would prefer low code options . Design the ingestion to support new year’s data to come in without having to change code.
2. The teams are willing to adapt CI/CD, how can that be made possible with this solution?
3. The end - result is to create dashboards showcasing trends across years, any exploratory analysis that can be highlighted from the data will help the schools to fine tune their policies. Think about the serving layer for the dashboards, ensure it is cost-effective and scalable, the team's inclination towards the cloud is because they know it is a pay as go model, PaaS services that can be paused when not in use.
4. The school is very concerned about security; how can security be embedded across the solution?
5. Please also share the performance and pricing details for the solution.

Dataset: Downloads | NYSED Data Site // Enrollment Database

Hint: Azure Synapse, Power BI

 

What to Submit:

Provide a URL to your code repository for judging and testing. If your repo is private, it must be shared with elizabete.kalnozola@girlsintech.org before the deadline.

All of the following items must be included in your submission:

1. Submission form 
Submit your project to the Devpost before the deadline and complete the submission form questionnaire. Submissions (including video/code links) are editable until the deadline.
What to include: Problem Statement Partner. Introduction. Purpose & Motivation. How does the application work? How was the application developed? How to use the application? Difficulties & Challenges faced during the design and/or development process? Go-to-Market (How will the application be available to the public, and is it scalable?)

2. Video
Upload a 3-5 min video including the following: Problem Statement Partner. Pitch Deck Presentation. Demo of your product, including an explanation of the solution and function of your application. This video can be submitted as a Youtube link on your Devpost page for your project. Make sure the video is set to public. Note: Any video longer than 5 mins will be automatically disqualified

3. GitHub or other code repo link
Please ensure that this link is publicly shareable so judges can access your project

 4. Provide images, screenshots, & wireframes of your project

***Important: By submitting your project to the Code; Without Barriers Hackathon Devpost, you acknowledge that these materials may be used to promote this hackathon globally.