Georgi Urumov • 2023-02-12
Have your progress notes written for you automatically
OpenAI's Generative Pretrained Transformer (GPT) is a cutting-edge language model that uses deep learning techniques to construct human-like dialogues, answer questions, write stories, and perform a variety of natural language processing tasks.
GPT has an excellent capacity to recognize context and respond appropriately, resulting in extremely precise and polished output. The model is at the forefront of the area of language processing, leveraging its massive processing capacity to analyze and interpret huge datasets, therefore creating text that properly reflects the patterns and connections inherent within.
ChatGPT is the latest variation of GPT, developed by OpenAI that launched as a prototype on November 30, 2023, and was available for public use in December with 57 million monthly active users in its first month of availability and 100 million users two months after its release. It has been specifically trained for conversational interactions. This allows for real-time text generation in response to user inputs, making it ideal for chatbots, virtual assistants, and other conversational interfaces.
GPT-3.5, which was used for ChatGPT, was an improvement from GPT-3. There is a lot of hype and rumors about GPT-4 being 100x more powerful than GPT-3. But OpenAI’s CEO, Sam Altman denied it.
Due to their question-answering, summarization, and analysis capabilities; GPT-based models can be utilized with varied success as an assistant for some tasks in the therapist’s workflow - summarizing the content of a conversation, identifying the main topics, or even writing a passable SOAP note (provided information about the session). Still, there are some major limitations that users have to overcome.
The most obvious limitation of all GPT models is that they can only work with raw text data as input - no CSVs, JSONs, Word documents, PDFs, audio, or video files. This restriction complicates the use of the models and results in a substantial overhead for handling different data types.
Additionally, the maximum input size is relatively small compared to the average length of a therapy session. Solving this introduces a bunch of technical challenges and makes it unsuitable for the general user.
Due to the complexity of GPT models and their “Black box” nature, it is impossible to consistently get good output, without spending significant time figuring out the correct parameters and queries.
Our experience shows that sometimes a single question mark or a couple of words can make a world of difference. The process of finding the right model parameters and prompts is such a challenge that it has spawned a whole new field of study - Prompt Engineering
OpenAI’s GPT models are many things but one - HIPAA-compliant. The data which you submit to their APIs and models can be stored and used in the future for activities such as training on the next iterations of the GPT models.
Despite OpenAI’s rigorous process of verifying the ethical and correct usage of their models, the missing HIPAA compliance means that sensitive information cannot be uploaded without first removing all Personally Identifiable Information. This presents another big challenge for most of their users.
GPT-type models have been trained on huge text-based corpora including a tremendous variety of topics - from children's books to astrophysics textbooks. Having this variety makes the models extremely capable of knowing a bit about everything. Sadly, it results in an obvious lack of deep understanding of any given topic, especially one as complicated as psychology.
In order to mitigate this, one must fine-tune the model on carefully curated data and design intricate processes to make up for the lacking depth. This is not only difficult but also quite pricey.
Another thing that GPT models naturally lack is the capability for maintaining long-term context. As there is no inbuild state or permanent storage, the information contained in previous inputs is unavailable (GPT) or forgotten (ChatGPT) over time. The important task of maintaining context, critical for the correct conceptualization of therapy notes, is left to the user.
The good news is that despite all of those challenges, there are ways to successfully utilize GPT-type models. Let us show you how!
Mentalyc is an AI scribe designed specifically for psychotherapists that leverages the latest NLP techniques, including GPT models. It saves a lot of time by converting recorded audio sessions into detailed individualized notes. That way there is no need to take notes by yourself!
Mentalyc has a minimal to no setup and works based on recorded audio, which it processes to create notes in a flash. You only have to record the session and upload the file!
In simple words, the Mentalyc AI note-taking method converts your session audio into a verbatim transcript and a fully editable psychotherapy progress note. It ensures complete anonymity and HIPAA compliance, as no client data is required to create the notes.
At Mentalyc, it took us months of research and experimentation to consistently extract clinically relevant information from therapy recordings. We developed various approaches, models, and intricate data pipelines, which combine our in-house solutions with the intelligence provided by different GPTs, enabling a one-of-the-kind solution for generating different types of psychotherapy notes.
Another crucial aspect is the support our sophisticated AI can provide by generating statistics, detecting various symptoms, and even performing OFAID analysis (onset, frequency, and intensity of symptoms). When there is insufficient information, the AI scribe can offer an evaluation of the primary concern and any contributing factors which could be used to investigate further during an upcoming session with the client.
Mental health is a sensitive topic that requires strict adherence to regulations, particularly the Health Insurance Portability and Accountability Act (HIPAA), which governs the protection and privacy of medical records and health information. Other summarization tools in the market may not meet the standards of HIPAA compliance, making them unsuitable for use in the mental health industry. Additionally, their output is limited to a simple summary of the meeting, lacking the depth and detail necessary for a clinical note.
In contrast, Mentalyc app prioritizes HIPAA compliance, ensuring the protection and privacy of the client's health information. The company has invested time and resources into researching and experimenting with various techniques and models to provide its users with the final clinical note.
We continuously analyze and improve our system by rigorously tracking changes therapists make to their notes and measuring their satisfaction levels. Our system improves daily as a result of this ongoing research.
Integration of Electronic Health Records (EHRs) is a critical component in providing seamless healthcare services. Mentalyc recognizes the need of connecting with existing EHR systems in order to provide a complete solution for healthcare professionals. That is why we have made our notes API-accessible, allowing for rapid and easy integration with other EHRs.
Healthcare practitioners may access and use Mentalyc's notes inside their existing systems with just a few clicks, expediting the process and removing the need for manual data entry. This integration keeps all patient information in one place, allowing for a full and structured picture of each patient's health history. Integration of Mentalyc's notes with other EHRs results in better patient care, more productivity, and less administrative stress for clinicians.
At Mentalyc, we understand the importance of data security and have made it a top priority when developing our tool. To ensure the safety and confidentiality of our user's information, we store all data entered into Mentalyc in HIPAA-compliant databases that are fully encrypted, making it inaccessible to even our own team. The app also automatically deletes entire session recordings once the note has been created, providing an extra layer of protection for our users.
Before starting to use Mentalyc in your practice, please note that the client's consent is required to be able to use session recordings. To make it easier for the therapists, we have created a downloadable document you can send to your clients to gain their consent before using Mentalyc.
To match the standard of our notes, you would need a team of Machine learning engineers, psychotherapists, data scientists, and developers, along with a way to collaborate and share knowledge. Additionally, you would need a legally collected and labelled dataset of therapy recordings for fine-tuning and improving the results.
Also, mental health data is sensitive and regulated by HIPPA. Building a compliant solution from scratch can be difficult and requires a thorough understanding of privacy laws and regulations.
Choosing Mentalyc can save you time and money as you won't have to reinvent the wheel. Additionally, you will have access to an expanding variety of note types and case analyses over time as we work on connecting data points between sessions.
Even large tech providers face obstacles in replicating Mentalyc's intricate solution, and a smaller private practice would simply not have the needed resources and know-how. With Mentalyc, you can simply drag and drop your recording and get notes.
Mentalyc is a perfect tool to take away manual note-taking from your practice and allow you to fully focus on your sessions without stressing about documentation. Mentalyc offers you easy-to-use solutions for your note-taking needs at super-affordable rates.
All you have to do is to take an audio recording from your session and upload it to Mentalyc. You’ll have your notes ready in no time, with a progress note and verbatim transcript. Mentalyc API can also be integrated with existing EHR systems, taking away the hassle of manual data entry.
If you’re an individual practitioner or a practice owner of any size trying to create progress notes for your sessions fast and with consistent accuracy, then Mentalyc is the best tool you can get.
No, AI cannot replace human counselors because it lacks empathy and emotional intelligence, both of which are essential in therapy. While AI might help with jobs like data analysis, summarization, organization, and personalization, human interaction is still vital.
Artificial intelligence is being applied in mental health in a variety of ways, including virtual treatment and coaching, note-taking and note organization, mood tracking, and symptom analysis. Although AI-powered technologies can assist mental health practitioners with diagnosis and treatment planning, human therapists still conduct therapy sessions.
Yes, AI in the form of chatbots and virtual assistants may be used for treatment, but it is not designed to replace human therapists and still requires the supervision of mental health specialists.
CSV - Comma Separated Values
EHR - Electronic Health Records
EMDR - Eye Movement Desensitization and Reprocessing
Georgi is a highly skilled professional with more than a decade of experience as a full-stack developer and Machine Learning Engineer. He has an extensive background in diverse fields such as process automation, web development, Web3, Computer Vision, and NLP. As the CTO of Mentalyc, a company he co-founded with Maria Szandrach in 2021, Georgi is committed to realizing his dream of using data-driven technologies to create human-centric solutions that can help alleviate mental health issues.
All examples of mental health documentation are fictional and for informational purposes only.
Speed up your note-taking process
Copyright © 2023 Mentalyc Inc. All rights reserved.
Who we serve