Evaluation
Evaluation of the Project
The project has delivered several core functionalities that are crucial for clinical settings, yet it also presents areas for improvement. Below is a detailed analysis of its strengths and areas for enhancement:
User Interface
Strengths:
- The interface is simple and adheres to the smart glasses theme, offering an uncluttered experience that minimizes distractions during clinical use.
Areas for Improvement:
- Although the simplicity aligns with the design ethos, it may be too minimalistic in certain contexts. Enhancing visual feedback and interactive guidance could help users navigate the system more intuitively.
Functionality
Strengths:
- Audio Recording: The system effectively handles essential recording functions—start, stop, and pause—while seamlessly transferring recordings to the device, ensuring that transcripts can be generated reliably.
- Transcription: It accurately converts speech to text offline, stores transcripts locally, and even differentiates speakers through diarization. Support for multiple languages further broadens its applicability.
- Summarisation: The integration of a large language model to generate concise clinical reports from raw transcripts has been a significant asset, improving documentation efficiency.
Areas for Improvement:
- Enhanced Hardware Integration: Currently, the smart glasses primarily serve as a microphone. Leveraging additional features—such as the built-in camera or augmented reality capabilities—could provide a more comprehensive user experience.
- Expanded Features: Introducing functionalities like manual uploading of audio files, as well as options to download transcripts and reports, would enhance versatility and user control.
- More Robust Chunking: More checks to ensure that a text is in a specific format and chunk accordingly. If the text given is not in the required format, the chunking doesn’t work properly, potentially resulting in an overflow of the context window.
- AI Context Window: For future use cases with very large documents or long conversations, consider extending the project to work with multiple LLMs to handle summarisation collectively.
Stability
Strengths:
- The application is generally crash-resistant under standard operating conditions.
Areas for Improvement:
- Stability concerns arise with longer audio files, especially on less powerful machines. Optimizing the system to handle extended recordings or segmenting long audio inputs may mitigate these issues.
Efficiency
Strengths:
- The offline processing approach ensures data privacy and security.
Areas for Improvement:
- The performance of the underlying models is heavily dependent on having a high-performance computer. This requirement limits accessibility for some users, as lower-end systems lead to slower processing speeds. Future iterations could benefit from model optimizations or the option for cloud-based processing as an alternative.
Compatibility
Strengths:
- The project has successfully integrated key functionalities needed for clinical environments.
Areas for Improvement:
- There is room for improvement in terms of compatibility with FHIR (Fast Healthcare Interoperability Resources) standards. Enhancing interoperability would facilitate smoother integration with existing healthcare systems.
Maintainability
Strengths:
- The project is built on a well-defined set of functionalities with clear modular objectives.
Areas for Improvement:
- Reliance on numerous external dependencies (such as OpenVINO, various npm packages, and Python libraries) introduces risks associated with future deprecation. Adopting a more modular codebase, comprehensive documentation, and regular dependency audits could greatly enhance long-term maintainability.
Project Management
Strengths:
- Regular check-ins with the client ensured that project deliverables aligned with user expectations.
Areas for Improvement:
- Early-stage communication was challenged by the involvement of multiple stakeholders, leading to coordination difficulties. More frequent progress updates and establishing clear communication protocols from the outset could have further streamlined project execution.
Future Work
Given that this project is a proof of concept, future work could significantly expand its functionality and robustness by integrating additional smart glasses features such as the camera and augmented reality to move beyond the current microphone-only interface. Exploring cloud integration could boost efficiency on less powerful hardware, although data protection would need careful consideration. We could also work on ways to improve summarisation for large context windows as an extension of the project.
Website Contribution |
|
|
|
|
Work Packages |
Annaelle Mansuy |
Deependra Datta |
Cici Liu |
Yao Ming |
Website Template and Setup |
0 |
0 |
100 |
0 |
Home |
35 |
20 |
20 |
25 |
Video |
20 |
20 |
50 |
10 |
Requirements |
20 |
50 |
25 |
5 |
Research |
40 |
10 |
10 |
40 |
UI Design |
10 |
90 |
0 |
0 |
System Design |
100 |
0 |
0 |
0 |
Implementation |
25 |
25 |
25 |
25 |
Testing |
0 |
0 |
0 |
100 |
Evaluation and Future Work |
10 |
70 |
10 |
10 |
User and Deployment Manuals |
20 |
80 |
0 |
0 |
Legal Issues |
0 |
80 |
0 |
20 |
Blog and Monthly Video |
25 |
25 |
25 |
25 |
Overall contribution |
25 |
25 |
25 |
25 |
Index |
Feature |
Sub-Index |
Requirements |
Category |
State |
Contributers |
1 |
Audio Recording |
1.1 |
Start, stop and pause audio recording |
Must Have |
Complete |
Cici, Annaelle |
|
|
1.2 |
Transfer over the recordings to your device |
Must Have |
Complete |
Cici,Ming |
2 |
Transcription |
2.1 |
Convert speech to text |
Must Have |
Complete |
Ming, Annaelle |
|
|
2.2 |
Store transcripts of the coversation on your device |
Must Have |
Complete |
Ming, Annaelle |
|
|
2.3 |
Speaker diarization |
Should Have |
Complete |
Ming, Annaelle |
3 |
Summarisation |
3.1 |
Summarise transcript with use LLM |
Must Have |
Complete |
Annaelle |
|
|
3.2 |
Fact Check |
Could Have |
Complete |
Annaelle |
4 |
User Interaction |
4.1 |
Access information through an application |
Must Have |
Complete |
Deependra, Annaelle |
|
|
4.2 |
Ability to edit reports and transcriptions |
Should Have |
Complete |
Deependra |
|
|
4.3 |
Ability to create an account |
Should Have |
Complete |
Deependra |
|
|
4.4 |
Ability to add patients |
Should Have |
Complete |
Deependra |
5 |
Integration |
5.1 |
Deployment |
Must Have |
Not Complete |
Ming |
|
|
5.2 |
NPU compatibility |
Could Have |
Semi-complete |
Cici |
|
|
5.3 |
Intel OpenVINO |
Should Have |
Complete |
Ming, Annaelle |
6 |
Security |
6.1 |
Login and logout |
Must Have |
Complete |
Deependra |
|
|
6.2 |
Password Change |
Should Have |
Complete |
Deependra |
|
|
6.3 |
Processed Locally |
Must Have |
Complete |
All |
Index |
Bug Description |
Priority |
1 |
Chunking not robust. |
Medium |
2 |
Tapping glasses too fast can result in synchronisation errors |
Low |
3 |
Parsing not robust |
Medium |
4 |
Packaged product causes memory issues |
High |
System Artefacts Contribution |
|
|
|
|
Work Packages |
Annaelle Mansuy |
Deependra Datta |
Cici Liu |
Yao Ming |
Initial designs (HCI) |
25 |
25 |
25 |
25 |
Research |
25 |
25 |
25 |
25 |
Experiments |
55 |
15 |
15 |
15 |
Implement Backend |
40 |
20 |
20 |
20 |
Implement Frontend |
20 |
80 |
0 |
0 |
Finish 'Must Have' |
25 |
25 |
25 |
25 |
Finish 'Should Have' |
25 |
40 |
10 |
25 |
Finish 'Could Have' |
60 |
0 |
40 |
0 |
Test Product |
3 |
3 |
3 |
91 |
Overall Contribution |
25% |
25% |
25% |
25% |