VITALS

Presentation
Introduction
Uniqueness
Problem Statement
Proposed Solution
Impact and Target Audience
Feasibility and Challenges faced and how we overcame it
Thank You

Call the Live AI Voice Agent now

+1 657-332-0285

Inspiration

Current healthcare is largely reactive and manual. We were motivated by the fact that patients often ignore subtle symptom links, such as a metallic taste in tea, and that doctors lose up to 70% of consultation time to repetitive history-taking. We built VITALS to bridge the care gap between overburdened clinical staff and chronic patients, aligning with SDG 3 (Health and Well-being).

What it does

VITALS is a RAG-powered multilingual AI voice agent designed to detect early illness escalation.

It conducts automated, human-like check-in calls to monitor chronic patients autonomously.
The system uses a Symptom-Escalation RAG model to cross-reference patient responses with historical data to flag high-risk anomalies.
It asks dynamic, context-aware questions based on the patient's specific chronic condition rather than following a script.
Summarized clinical transcripts and reports are generated for doctor approval.
Upon approval, actions like dosage changes or emergency appointments are automatically sent to the patient via WhatsApp.

How we built it

We utilized a cutting-edge AI stack to ensure a seamless and professional experience:

Voice & Telephony: Built using Twilio and Vapi to bridge AI agents with the telephony system.
Human-like Pipeline: Integrated Deepgram for Speech-to-Text and ElevenLabs for indistinguishable Text-to-Speech.
Intelligence: Leveraged Retrieval-Augmented Generation (RAG) to prevent hallucinations and maintain source transparency.
Privacy: Developed a De-identification layer to strip patient identities before data hits the LLM.

Challenges we ran into

Medical Accuracy: Addressing potential "Medical Hallucinations" required a strict RAG-grounded approach and keeping a doctor-in-the-loop for final safety.
Latency: Reducing voice call delays required implementing low-latency WebSockets (Vapi/Retell) to maintain natural conversation speeds.
API Costs: Managing high costs for specialized models necessitated strategies like model distillation and the use of smaller, specialized LLMs.

Accomplishments that we're proud of

Human-Like Interaction: Successfully engineered an AI voice pipeline that is indistinguishable from a human voice, fostering the trust necessary for chronic patient monitoring.
Clinical Time Recovery: Developed a system that automates the initial history-taking process, which typically consumes up to 70% of a doctor's consultation time.
Preventive Healthcare: Implemented a RAG-driven detection system that identifies high-risk anomalies early, helping to prevent the progression of chronic illnesses into acute emergencies.
Staff Burnout Reduction: Designed an automated outreach framework capable of handling repetitive follow-up calls, reducing the administrative monitoring burden on nursing staff by an estimated 65%.
Scalable Monitoring: Built an infrastructure that allows for monitoring hundreds of patients simultaneously, providing a level of proactive care that is impossible for manual staff alone.

What we learned

We learned that Source Transparency is critical in health-tech; unlike competitors that may lack grounded results, our RAG-enhanced approach provides the clinical-grade accuracy needed for diagnosis. We also gained experience in overcoming legal and privacy hurdles through the implementation of a de-identification layer and private vector databases.

What's next for VITALS

Our roadmap includes deeper integration with existing EHR systems to reduce deployment capital by 40%. We also plan to expand our autonomous escalation capabilities to better handle recovery trends and continue developing proactive support for chronic populations globally.

Experience the live prototype: Call +1 657-332-0285