CTS NPN Hackathon Team 13 Use Case Discussion

Main ideas, concepts, and lessons conveyed

• The video presents a hackathon team’s healthcare project called “AA” (Patient/Doctor “assistant”), focused on improving patient experience in hospitals.
• Problem addressed: Many hospitals/patient desk assistants lack strong patient experience, and competition among healthcare providers increases the need to retain patients.
• Core concept of the solution: Enhance existing patient-facing services with intelligent, personalized automation, including:
  • information support
  • record handling
  • discharge reporting
  • consultation timing prediction
  • eligibility checks for government schemes

System purpose and objectives (as described)

• Appointments: support and automation for booking/handling appointment-related tasks.
• Displaying reports: provide access to patient reports.
• Sharing patient history: retrieve and present patient history when needed.
• Automation focus: include only features useful/necessary for automation, emphasizing improved personalization and intelligence over existing solutions.

Feature list / capabilities of “AA” (organized from the subtitles)

Patient communication & support

• General inquiry chatbot

  • Answers common questions using a rule-based chatbot flow
  • Provides personalized automated feedback

• Follow-up and messaging after treatment

  • Reminds patients about:
    • medication
    • appointments
    • self-care tips
  • Aims to improve recovery and reduce likelihood of complications by keeping patients informed during recovery.

Medication and consultation assistance

• Alerts to patients
  • Medication reminders
  • Consultation-related prompts/alerts

Patient data support

• Retrieve patient records/history
  • Provides access to:
    • previous treatments
    • medications
    • outcomes
  • Intended to help track progress over time and support more informed decision-making.

Consultation time optimization

• Predict optimal consultation timing
  • In heavy/busy hospital situations, predicts the best time for the patient to meet the doctor to finish consultation efficiently.

Discharge report generation

• Smart discharge report generation
  • Produces a patient-friendly report by combining:
    • relevant visit details
    • symptoms
    • doctor recommendations

Government scheme eligibility checking

• Government scheme eligibility check
  • Determines eligibility for common schemes such as:
    • insurance-related coverage
    • healthcare services
  • Based on patient information (including personal and medicine-related information).

Rule-based chatbot workflow / “tree diagram” logic

• Step 1: Greeting

  • The chatbot greets the user.

• Step 2: Query category selection

  • The chatbot asks what type of query the user has, with examples including:
    • General information
    • Doctor availability
    • Visitor information
    • Contact information

• Step 3: Handle “General information” queries

  • If the user selects general information, the chatbot prompts for a specific question such as:
    • Department services
    • Hospital hours
    • Location

• Step 4: Example responses by selected option

  • If the user selects Hospital hours:
    • provides opening/closing times and/or services offered
  • If the user selects Location:
    • provides the hospital location

• Step 5: Repeat coverage

  • The chatbot continues handling queries to answer each information request (according to predefined rules).

Discharge report generation methodology / workflow

• Goal

  • Efficiently generate a patient-friendly discharge report by pulling data from multiple tables and combining:
    • patient visit details
    • symptoms
    • doctor recommendations/instructions

• Step-by-step workflow

  • Input: Receive a visit ID

    • Use the visit ID to fetch relevant information from different data tables.

  • Fetch patient information

    • Query the patient table for:
      • name, age, medical history

  • Fetch appointment/visit details

    • Query the appointment history table for:
      • date of visit
      • cause of visit
      • symptoms
      • medications prescribed by the doctor

  • Fetch doctor information

    • Query the doctor table for:
      • doctor name
      • doctor ID

  • Generate report

    • Combine gathered information into a detailed report including:
      • reasoning behind the patient’s condition
      • prescribed instructions (doctor instructions)

  • Output

    • Present the generated report so the patient can understand their health better.

• Integration rationale (as stated)

  • Pulling from multiple sources makes the report accurate, complete, and patient-friendly.

AI model integration + predictive model methodology

LLM/discharge report generation integration (Gemini-based)

• Step 1: Install required libraries

  • Set up dependencies needed for Gemini integration.

• Step 2: Configure API key

  • Provide an API key to authenticate access to the Google AI platform.

• Step 3: Set up the generative model

  • Adjust generation parameters:
    • temperature
    • top p
    • top k
    • max output tokens
  • Purpose: produce responses that are accurate, detailed, and suitable for medical reports.

• Step 4: Add system instructions

  • Provide concise and precise instructions guiding output formatting.

• Step 5: Provide patient data to the model

  • Start a chat session (initially empty).
  • Feed the patient/visit data into the chat session to generate the discharge report.

Random Forest model for consultation time prediction

• Model purpose

  • Predict optimal consultation time for appointment scheduling.

• Data preprocessing

  • Columns:
    • appointment start time
    • optimal consultation time
  • Convert time from HH:MM format to total minutes to enable numeric operations for model training.

• Input features used

  • total slots
  • total available appointment slots
  • number of booked slots before the patient’s slot
  • number of empty slots before the patient’s slot
  • average consulting time
  • appointment start time

• Training/testing split

  • Use an 80/20 split for training/testing to evaluate performance.

• Model choice

  • Use RandomForestRegressor with:
    • 100 estimators

• Evaluation metrics mentioned

  • Mean Absolute Error (MAE): average magnitude of errors
  • Mean Squared Error (MSE): emphasizes larger errors
  • R-squared (R²): proportion of variance explained by the features

• Stated expected benefit

  • Improve scheduling by reducing waiting time, improving operational efficiency, and optimizing consultation timing.

Speakers / sources featured (identified from subtitles)

• Speaker 1: “Tarun” (team member; also referenced during the intro as part of Team 1)
• Speaker 2: “K” (teammate; explained features)
• Speaker 3: “Deepak” (continued presentation with features; introduced discharge report workflow)
• Speaker 4: “Gemini / Google Generative AI” (source/tool used; not a human speaker)