TheraCaIl

Logo
One flowchart for project design
Flowchart for the LLM dynamic script
Website landing page

Inspiration

One of our members is currently a therapist at Kaiser and a UC Berkeley Master's student. She has worked with about 100 patients, who meet weekly or bi-weekly, and she's increasingly worried. Therapy isn't about what happens in the room; it’s about what happens in the life of the patient between sessions. In Cognitive Behavioral Therapy, they call this an 'Action Plan.' It’s the engine of recovery. However, there is a massive gap in the continuum of care. Between sessions, they are left entirely unsupported for over 300 hours. Research indicates that up to 50% of patients are unable to follow their treatment plans without inter-session support consistently. And when the 'homework' fails, the therapy fails.

What it does

Our website is a tool for therapists to be able to quickly see their patients' status while they receive an AI phone call.

AI-Powered Check-ins: The system conducts automated clinical calls using a 5-part protocol to check patient info, depression/anxiety levels, safety risks, and action plan compliance.
Risk Detection: It identifies high-risk scenarios, such as suicidal ideation, and provides immediate safety warnings to clinicians.
Action Plan Support: It follows up on specific therapeutic homework, such as Progressive Muscle Relaxation (PMR) or thought journaling, helping patients stay on track.
Clinician Dashboard: Therapists like our own teammate Emily can view a comprehensive dashboard that categorizes patients by risk level (Low, Medium, High) and highlights urgent actions required.

How we built it

We developed TheraCaIl using a powerful tech-stack to handle real-time voice communication and data processing.

Frontend: Built with React in order to provide a user-friendly and intuitive experience.
Backend: Built a custom API to help us bring our vision to life by weaving together many different APIs and libraries.
AI and Voice: We used Eleven Labs along with Twilio to work with live interactions over the phone, while keeping it empathetic and caring.
Database: Used MongoDB as our primary data management to store our patients information and update the website live after the phone calls.
Call Summary: We implemented Gemini API to grab our users' responses and generate a summary for the clinician to read.

Challenges we ran into

There were many challenges when developing TheraCaIl, especially because we used a lot of technology that we were new to.

Backend: Our backend was one of the most challenging parts of our project, using this many new API's was definitely a hurdle that took us a long time to overcome.
Connection: Connecting our frontend to our backend was also a challenge because of the limited time we had. We were constantly implementing new features, causing large mismatches of versions, and is definitely a learning experience.
ElevenLabs & MongoDB: When implementing our AI agent, we were having trouble with saving patient data to our database because of differing variables. It was a simple mistake, but it cost us lots of time.
Webhooks: Using webhooks was difficult for us because we were having connectivity issues once implementing our voice-over-phone call.

Accomplishments that we're proud of

Overall, this was a great project and learning experience, and there was a lot to be proud of.

ElevenLabs: Once we configured ElevenAgents with a dynamic script and user input we felt like we really got the ball going and it felt like a major success.
Hooks: Using the "useEffect" hook in React for the live AI-summary is what really makes our UI stand out.

What we learned

This project taught us that when patients fail to execute their action plans, they often mistakenly conclude that 'therapy doesn't work,' leading to worsened outcomes or dropped treatment.
We learned that while artificial intelligence cannot replace the human empathy of a trained provider, it is the perfect tool to solve this specific logistical failure.
By building a system that automates the follow-up process, we saw how technology can bridge the gap between sessions, providing the low-friction accountability patients need without adding to a provider's overwhelming workload.
Additionally, by working on developing this project, we also gained deeper insights into integrating multiple APIs (Twilio, ElevenLabs, Gemini) to create a seamless, voice-first user experience.

What's next for TheraCaIl

There were lots of features that we wanted to implement, but we didn't have the time.

AI Helper: We wanted to include a helper that would allow the therapist to create an action plan efficiently with only a few points of entry.
Condition Levels: We also wanted to implement a real-time table that would update the provider on the patient's condition based on the checkup call.