Healthcare worker assisting a mother and baby in a clinic
Healthcare

Jacaranda Health's PROMPTS AI Triages SMS Messages to Save Mothers and Newborns Across Kenya

7 min read|Updated March 2026
Share

At a glance

Jacaranda Health created PROMPTS, an AI enabled SMS platform that uses natural language processing and fine tuned local language models like UlizaLlama to answer maternal health questions and triage over 5,000 daily messages from pregnant women across Kenya. Serving more than three million mothers at less than 75 cents per user, the system identifies life threatening emergency signs within minutes and connects high risk mothers directly to clinical helpdesks and hospital facilities.

  • Maternal mortality in sub Saharan Africa remains high, with care seeking delays contributing to one third of maternal deaths in Kenya
  • PROMPTS delivers stage specific pregnancy and postpartum SMS guidance in English, Kiswahili, and local dialects
  • Natural language processing models triage 5,000 daily incoming user messages, automatically flagging emergency symptoms like severe bleeding or preeclampsia
  • Jacaranda fine tuned Meta's Llama 2 model into UlizaLlama to accurately decode code switching between Swahili, English, and local Sheng dialects
  • The platform has reached over three million mothers across 1,100 health facilities at an operational cost under 75 cents per participant

Maternal mortality remains a critical global health challenge, with sub Saharan Africa accounting for roughly 70 percent of maternal deaths worldwide. In Kenya, operational delays in identifying danger signs and seeking emergency care contribute directly to one third of all maternal deaths. Jacaranda Health, a non profit healthcare organization operating across East Africa, developed an artificial intelligence platform called PROMPTS to address this gap. Operating over low cost SMS text messages, PROMPTS delivers personalized clinical advice to pregnant women and uses natural language processing to triage thousands of incoming health queries every day.

By using advanced language algorithms fine tuned on regional communication patterns, Jacaranda Health has scaled PROMPTS to reach over three million women across Kenya. The system operates at a lifetime cost of less than 75 cents per mother. By combining low tech mobile SMS with open source machine learning, Jacaranda Health is reducing clinical response times from hours to minutes, ensuring that high risk obstetric complications receive immediate emergency medical intervention.

The Critical Gap in Maternal Care Seeking

In many developing health systems, expectant mothers face severe barriers when navigating pregnancy and postpartum recovery. Obstetric danger signs such as heavy bleeding, persistent severe headaches, high fever, or reduced fetal movement require immediate evaluation at a equipped medical facility. However, lack of maternal health literacy and uncertainty about symptom urgency frequently cause delays in seeking care.

Public health studies in Kenya reveal that 91 percent of women state they would seek immediate hospital care if they recognized severe pregnancy warning signs. The core obstacle is not a lack of willingness, but rather a lack of timely, accurate information tailored to the mother's specific stage of pregnancy. When expectant mothers rely on informal advice or wait until symptoms escalate, preventable complications like preeclampsia or hemorrhage can turn fatal before reaching a hospital.

How AI Triages 5,000 Messages Every Day

PROMPTS functions as an automated digital companion throughout pregnancy and the postpartum period. Enrolled mothers receive weekly SMS text messages tailored to their exact gestational week. These messages explain fetal development, remind mothers of upcoming antenatal care checkups, and educate them on critical health red flags.

Because mothers can reply directly to any SMS message with questions about their symptoms, Jacaranda Health receives over 5,000 incoming text messages every day. Human clinical helpdesk agents cannot manually read and prioritize thousands of daily messages fast enough to catch urgent medical emergencies. To solve this bottleneck, Jacaranda Health developed TRIM AI, a specialized natural language processing architecture designed to tele triage incoming texts in real time.

The AI platform analyzes incoming text messages and assigns an urgency score based on clinical risk algorithms. Messages indicating severe complications like severe abdominal pain, fluid leakage, or fever are instantly categorized as high priority red flags. The platform routes these priority cases to human nurses and helpdesk agents within seconds, while routine questions regarding diet or general discomfort receive automated responses or lower priority queueing.

UlizaLlama: Solving the Code Switching Challenge

Building an AI triage system for East Africa required overcoming significant linguistic hurdles. Expectant mothers in Kenya rarely write in standardized textbook English or formal Kiswahili. Instead, messages often feature fluid code switching between Swahili, English, and Sheng, an urban street slang popular among youth. Furthermore, texts frequently contain informal phonetic spellings, local colloquialisms, and custom emojis.

Off the shelf large language models struggled to interpret these mixed language messages, leading to misclassifications of critical health queries. To resolve this, Jacaranda Health adapted Meta's open source Llama 2 model to build UlizaLlama, a custom Swahili centered language model trained on millions of anonymized real world health messages from Kenyan mothers.

In field evaluations, UlizaLlama demonstrated a 17 percent increase in diagnostic accuracy when identifying high risk medical conditions from informal SMS texts compared to baseline models. This artificial intelligence breakthrough reduced helpdesk processing workloads by 12 percent and ensured that emergency clinical signals were recognized regardless of how the mother phrased her message.

PULSE: Tracking Healthcare Facility Quality in Real Time

Beyond direct maternal communication, PROMPTS acts as a large scale health system monitor through its PULSE analytics framework. After attending antenatal visits at local clinics, mothers receive short survey prompts asking whether essential clinical protocols were completed during their visit, such as blood pressure screening, urine testing, or respectful treatment by clinic staff.

Jacaranda Health aggregates these anonymized survey responses into real time dashboards accessible by county health directors and hospital managers. If data reveals that blood pressure checks have dropped in a specific district, health officials can quickly investigate equipment shortages or staff training gaps. By transforming patient feedback into actionable health systems intelligence, PROMPTS improves clinical quality across more than 1,100 public health facilities.

Measurable Health Outcomes for Three Million Women

The operational scale and clinical efficacy of PROMPTS have made it one of the most successful digital health interventions in Africa. Randomized evaluations and public health data demonstrate that mothers enrolled in PROMPTS are twice as likely to complete four or more antenatal care visits compared to non enrolled peers. Postpartum family planning adoption has doubled, and emergency care seeking times for acute complications have dropped significantly.

To date, PROMPTS has served more than three million women across Kenya. Because the platform relies on basic SMS infrastructure accessible on basic feature phones, it reaches underserved rural populations without requiring internet access or high end smartphones. At a total cost of under 75 cents per user for lifetime platform access, government health departments can sustainably integrate PROMPTS into national public health budgets.

Expanding AI Driven Maternal Care Across Africa

Building on its success in Kenya, Jacaranda Health is expanding the PROMPTS AI platform to other African nations, including Ghana and Eswatini. By sharing open source datasets, open language models like UlizaLlama, and tele triage code with global health partners, Jacaranda Health is establishing a global open model for digital maternal health.

As artificial intelligence technology advances, combining localized language models with accessible mobile networks provides a path forward for maternal healthcare in resource constrained regions. By listening to mothers in their own language and identifying medical emergencies in real time, AI tools are helping safeguard future generations across the Global South.

Common Questions

What is Jacaranda Health's PROMPTS platform?

PROMPTS is an AI powered mobile health service developed by Jacaranda Health that sends stage specific pregnancy and postpartum guidance via SMS and uses natural language processing to triage over 5,000 daily health queries from mothers across Kenya.

How does the AI triage high risk pregnancy cases?

The platform uses natural language processing models (TRIM AI) and custom fine tuned language models (UlizaLlama) to scan incoming SMS messages for red flag symptoms like severe bleeding or high fever, escalating urgent cases to human clinical agents within seconds.

What is UlizaLlama and why was it created?

UlizaLlama is an open source Swahili language model built by Jacaranda Health by fine tuning Meta's Llama 2. It was created to accurately decode informal text messages featuring code switching between English, Swahili, and local Sheng dialects.

How many mothers has PROMPTS reached and what does it cost?

PROMPTS has reached over three million women across Kenya and extended to other African countries. The platform operates at a cost of less than 75 cents per mother for lifetime access on the system.

Does PROMPTS require a smartphone or internet connection?

No, PROMPTS operates via standard mobile SMS text messaging, making it fully accessible on basic feature phones without requiring internet data or a smartphone.