DocAssist AI

• 

  Login page - demo accounts created for testing

• 

  Doctor Portal - doctor can add visits for each patient and add case notes - leverages LLM to summarise visit notes + diagnoses insights

• 

  Patient Portal - populates with booked visit - visit chat prompt patient to provide further info

• 

  conversation in real time with patient (AI dynamic generated Qs) - questions and answers displayed after submission of all answers

• 

  After chat with smart intake chat for visit - data is saved and displayed as AI generated summary for patient to see

• 

  After visit intake form completed - info appears on doctor portal with AI generated summary and AI generated potential diagnoses

• 

  Pre visit summary for GP with diagnosis and summary

• 

  During consult doctor can generate notes and then add or subtract info - once saved this will save as post visit summary.

• 

  Once doctor submits it is non-editable case notes summary that can be viewed by the doctor at any time.

Inspiration

Our inspiration for this project comes from the common complaints we have heard surrounding the current GP consultation process and experience. Our solution aims to reduce rushed appointments, remove the one-size-fits-all approach to diagnosis, and empower the GP to focus on complex problem solving and empathetic communication instead of information gathering. We also see potential benefits for this solution in democratizing healthcare and disrupting any unknown biases the GP may be holding in a supportive manner.

What it does

• Our solution sends a “smart intake form” to the patient. This smart-form is a conversation with the Arctic LLM that guides the patient through a set of questions to further investigate their symptoms and gather relevant information.
• This conversation is then summarized by the model and sent to the GP’s case notes file that can be viewed prior to the consultation. The summary includes key symptom details, a bias disruptor that flags potential less common diagnoses, and further questions or analyses the GP may want to pursue during the in-person consultation. This eliminates the difference in experience that patients have with doctors that have varying experience and knowledge.
• If an audio transcript is available of the GP-patient consultation, it can be uploaded to the system for the automatic generation of case notes. These can then be reviewed and edited by the GP prior to uploading.

How we built it

We took a collaborative approach to research and design, including a market scan, desktop research around the problem set, and feasibility assessment. The code was developed through splitting the work and branching the repository in GitHub. ChatGPT was useful in generating many first drafts of the code, especially for the components that we were less familiar with.

Challenges we ran into

• Prompting the Arctic LLM to prevent the duplication of responses when consulting the patient.
• Adjusting to specific features of the Streamlit library such as st.session_state.
• Completing the hackathon while working full time.

Accomplishments that we're proud of

• Having a functional MVP of a smart case notes system that can support the end-to-end GP consultation experience.
• Contributing to the health practice that we all depend on as individuals, even if it is at the conceptual stage of implementation.

What we learned

• Front-end Streamlit development.
• Accelerated prototyping of code.

What's next for DocAssist AI

• Conducting customer interviews with consumers and providers of medical services to gather further information and assess product-market fit.
• Connecting to other databases like weather/air quality.
• Implementing patient scheduling and follow-up AI conversations.
• Using RAG to fetch standards by country.
• Inputting different types of media like images, scans, audio, video, etc.
• Providing richer analysis.
• Persisting states so that refreshes do not require the user to re-log into the app
• Taking in different types of input and making it multi-modal

Built With

• anaconda
• arctic
• llm
• python
• snowflake
• streamlit