"Develop an in-app tool to allow setting up in-app nudges for various use cases, tool to allow nudge configuration on screen without need of having to code them on the screen / per use case."

How a nudge should get trigger:
Static Nudge: Placement of static nudges on the BFL app / Sample app sections. These are admin configured nudges done through front end without need to code on the app.
Automated event-based Nudge: Some nudges to trigger basis sample events / API or data set (tags) in a real-time manner. (Speed of visual trigger is essence)

Evaluation/Acceptance criterion:
Metric - We need proper metrics in place with backend logic with regards to the conversion ratio i.e. success of the particular nudge(s). Governance Control (Optional): We need a governance control to be available as per below:    
  1. Priority set: Setting of business product wise prioritization.
  2. Customer threshold: X no. of nudges to be triggered per customer (daily/Weekly/Month).

“Explore the data set shared (SNAP product co-purchasing networks). The key task is to first identify a good problem to solve that will address some potential business use case (you need to identify this). This solution needs to include a graph-based approach with the steps identified in the submission/evaluation section. This hackathon problem will explore on building scalable graph-based capability with a technical frame using big data, data pre and post processing, data models for Graph/ML/analytics consumption. The data is based on the data based on a large ecommerce platform and hence some use cases linked to that can also be considered”.

Outcomes Expected:
- Post understanding the dataset (both transaction and meta data) identify a problem that you would want to solve. Clearly articulate the problem statement.
- Data processing (including Loading and transformations) of the data set (combination of pyspark & python-based code) or other high-performance libraries (in case being used, need to mention with details)
- The data processing/analysis needs to use both transaction and meta-data(semi/(un)structured)
- Creation of intermediate and base analytical data models (using both transaction and meta data)
- Create the graph-based model to help solve your problem statement basis the standard libraries mentioned; in case you have a better library, please call-out purpose of using them including providing specifics on them.
- In case the models require post process for computation; please factor the needful details on how they have been used.

1	Data Set Details Product co-purchasing networks - SNAP)	Data	Code Notebook	Google collab Link
2	Data Set: Meta Data	Data	Languages & Framework	Python, Pyspark | Spark 2.4+
3	Data Set: To Use	Data	Graph Components	networkx | SNAP - Stanford | graphframes
3	Visualisation Tools	Data	NA	NA

Submission & Evaluation Broad Metrics (using Technical Stack)
- Problem statement definition (well-defined)
- Pre-processing and data cleaning (including standardising)
- Building reference/base data sets that facilitate modelling
- Exploratory Data Analysis (linked to problem statement)
- Data Processing pipelines
- Model Build Approach & Selection (includes testing) (using Graph based approach)
- Model Execution |Results & relevance as per problem statement
- Integration Approach | (broad technical approach)
- Model Management and Performance Tracking Approach
- Guidance to Code level
- Inline | Code | Documentation | Markdowns
- Quality (modular, coding guidelines, etc.)
- optimisations and efficiency codified basis your assignment
- Algorithm/Time complexity, scalability & efficiency)
- Performance (time, memory metrics, etc.)

Submission Specifics
-Must be a notebook (runnable live) as per platforms indicated above.
-Notebook must be named: HealthRx30_Hackathon_June22_
-Additional Document (if required): HealthRx30_Hackathon_June22_.doc
-Additional aspects can also be covered in the notebook as markdown or additional document can also be considered.

“One – Stop Social Sentiment curation analysis site for Hospitals/Laboratories"

We live in a Golden Age of Information where, with product information and reviews available across several outlets, making informed choices is a piece of cake. The problem now becomes, due to the scattered nature of such information, consumers spend a lot of time trying to collate and comprehend this data across platforms (a problem statement that defines the existence of companies like Yelp, Mouth-shut, Trustpilot, etc).

Problem Statement:Participants are encouraged to create a portal, powered entirely by publicly available data, that would cater to a wide range of users looking for information regarding reviews of Hospitals/Laboratories etc., The portal is meant for patients choosing a certain hospital in their city for a medical procedure, or hospital administrators looking for insights into improving their services.

Deliverables:
Following deliverables are expected:
- Create a portal, that displays information about Hospitals/Labs - powered entirely by publicly available data, that would be useful to any interested user.
- The data could be reviews scraped from social media, review sites, and other public domains, and the display sections could be aggregations, sentiment analysis derived visuals, specialities/procedures that a Hospital is well known for, etc.
- ROHINI (Registry of Hospitals in Network of Insurance) is a registry of unique hospitals in the Health Insurers and Third Party Administrators (TPAs) network, in India.
- You can create your own sentiment analysis engine that can score the reviews collated from the websites

Evaluation/Acceptance criterion:
- Solution approach to problem statement in terms of scalability with respect to Hospitals covered.
- Automations of processes to get incremental data from websites.
- Innovation in Display UI and data usage – showing reviews & ratings (sentiment analysis)
- Number of data sources used for gathering information
- All tools and frameworks should be open-sourced. 

"Implement a provider (Hospital and Labs) invoice text extraction to identify provider (Hospital/Dr./laboratory) names & other critical information tags."

At Bajaj Finserv Health, one of our products is OPD health insurance (doctor consultation, lab tests) benefits. The patient can claim these using reimbursement process - where invoices/bills of hospital/doctor/lab are provided to us for claims and claim agent reviews and approves/rejects the claim.

Problem Statement:
The bills provided are either printed OR handwritten. Claim agents spend a lot of time in viewing and entering the information on the bill. We need a more efficient way of capturing this information in a structured manner where provider and treatments can be extracted– to eventually use it for analysis purposes & also detect fraud.

Deliverables:
Create an API that will
- Accept the document and return ‘all text parsed from the document’ as well as categorized text tags with values
- Also return structured text which can be classified(tagged) into below categories (where available in document)
- Date of bill
- Provider name
- Provider address
- Telephone
- Pin
- City
- Service names (might be multiple) – services taken by the patient
- Amounts for the service
- Total amount of bills
- The tags might not always be explicit – so you will need to derive tags basis intelligence (e.g. detect that something is a DATE)
- API should be able to extract – printed OR handwritten text
- Identify if the document was ‘doctored’ i.e. someone has digitally modified it after taking photo (this is potential fraud)

Evaluation/Acceptance criterion:
- # of Categories (provider name/date etc.) detected
- Printed & OCR capabilities showcased
- Accuracy of information captured
- Ability to capture ‘full text blobs’ & derive insights around it
- Innovation on how this ‘data’ extracted can be used to detect fraud
- Additional use cases of the extracted text

[Sample 10-15 bills will be provided] 

"Mobile Devices are ubiquitous and almost everyone carries one. Our lives now depend on it and data it carries reveal a lot about us. We would like to use data on cell phone to assess credit behavior of the user."

Problem Statement:
Build a mobile application that uses data on mobile phone to assess behavior profile of a user.
The data can be:
- Mobile applications installed (Appography) to interpret user profile
- Non-personal SMS received to identify payment overdue
- Battery life, call usage etc., to determine behavior characteristics

Evaluation criterion: 
- Type of variables & logic used to infer user behavior and credit profile
- Accuracy of measurement and calculation of credit score

"More and more people are shopping online for things that were only bought in person. We would like to offer a digital experience to bring more customer online for items that are still bought physically."

Problem Statement:
Create a Mobile or Web Application that enables uses to try various products using augmented reality before buying them.

The products can range from fashion clothing, personal accessories like jewellery, spectacles’, make-up etc., to home furnishings like furniture, show pieces, decorative lights etc. Both real video augmentation and static image augmentation can be used.

Evaluation/Acceptance criterion:
Innovative approach to make digital experience as close to reality as possible Usage of technology to bring this innovation

Build a customer (UI/UX) prototype app/feature that will track customer spend habits. Understand spends related to health habits across various categories like food, drinks, fitness memberships, gadgets, etc. Build a reward structure (positive/negative) basis category of spends.

Acceptance /Evaluation criterion:
- Innovation in gamification.
- How seamlessly it is stitched to the journeys.

Open platforms like Twitter, FB, Mouth shut, Quora etc are considered to be neutral platforms where people share their experience about products, services and brands. The objective is to harness these platforms to identify and generate leads and prospects by applying website scrapping, text analytics, AI and push filtered data to Analytics engine for further processing.

How can customers (and influencers) help each other to solve issues, guide each other thru community support model:
- Harness the power of customer collaboration to promote brand, recommend products, explain product details etc.
- Need a solution where our customers/interested parties/Influencers can drive these conversations, boost brand presence and promote desired behaviour with customers.

“While customer servicing is the default use case for voice and claims is the default use case for video, can we explore relevant use cases across the spectrum?“

Voice and video bots can help customers with engagement and partners on status updates. Apart from voice enabled interactions with partners and employees for feedback and testimonials, focus should be to enable voice on digital assets (website, app and customer/partner/employee portals) and video enabled based risk inspection / auto adjudication of risk reports.

Detail Description:
In the current scenario, 41.6 lakh customers come to us for servicing wherein 40% of these customers reach out to contact centres for resolution. We also have 45.3 lakh claims registered. With the growing customer base, we are spending close to 15 crs on contact centres. Our inbound contact centre contributes to 40.3% of customer queries while 57.4% servicing is via our digital assets. Voice / conversational AI can be very handy beyond conventional customer life cycle use cases and can be implemented @ scale across transactions / interactions (customers, employees, partners & beyond). Product / process requirements should meet the regulatory and legal framework as well.

Solution Opportunities:
The challenge here is to automate and introduce a voice strategy to reduce the current customer contact footprint from 40.3% to below 5% and move the same to chat, voice and video bots. We also require a voice solution on repeated customer contacts wherein 25% customers contact us twice. The solutioning team needs to find the best strategy of contact centre automation along with a human touch of - 95% automation of customer resolution via voice and video & 5% for exceptional handling with a human touch.

The Opportunity:
“Conversational AI – Voice and Video” is the future and at BAGIC, we practice bringing “The Future Is Now - Tomorrow Happened Yesterday!
- Making our contact centre and services more accessible with automated virtual agents that are available 24 hours a day.
- Intelligent voice bots can answer calls with zero delay & a target of 5 seconds as max hold time
- Making virtual more personal - with access to relevant data, voicebots recognise customers, understand history and preferences, and use AI to personalise interactions and multilingual interactions
- Build an industry first voice enablement platforms across all touch points [customer facing, IMD facing, employee facing]
- Create and deploy voice bots in a multi-omni channel environment.
- Opportunity to automate 2 million customer calls at the contact centre
- With a customer base of 10+ crores and adding 2 crore+ every year, an agent base of 50k+ & employee strength of 10k+,  BAGIC has the opportunity to implement voice @ scale.

"Online search tools are important and powerful tools to assist customers to consume content and make decision in their online journey. The power of search is that it enables the customer to reach to the right content in shortest possible time (time optimizers). However, it is well now that the search engines at time undermine the power of search resulting in less optimized performance for the customer. This is due to some form of bias due to query not being able to capture the correct keywords"

Objective: Our aim is to suggest/ recommend keywords that are more relevant to user’s search query. This algorithm will help us to improve search results

Evaluation/ Acceptance Criteria:
- Accuracy - % results that are relevant
- Business output – direct measurement of search query to online sale in the buy journey
Accuracy means - % of query results that have returned the truth and nothing else
Business output means – the transactional impact of search on online business

Data Used
You need to crawl the site https://www.bajajallianz.com/general-insurance.html
Along with it, we will also share the search trends and learning from Adobe Analytics and Google ads to align the search results with larger search intent of the customer in the public search domain.
Based on the above data sets and problem statement recommend the algorithm with will help to further optimize and enhance the economic output of search on the website