SOP-ULZ-001 · Draft v0.9 for MDH review · 21 October 2026

Safe, effective and standardised use of the Uliza AI chatbot for HIV/SRH services

To be reviewed in a virtual meeting with MDH and approved by the relevant institutional authorities within the first month of the contract.

1. Purpose & scope

This SOP sets out how Uliza is operated so that adolescents and young people (AYP) aged 15–24 at participating universities and secondary schools in Dodoma receive accurate, confidential HIV/SRH information and are referred and linked to HTS, HIV self-testing, condoms, PrEP, PEP, family planning, STI, mental health and GBV services. It applies to Xakin AI staff, MDH programme staff, counsellors, peer educators and facility focal persons using the platform.

2. Roles

• Xakin AI (vendor): platform operation, availability (99.99%), maintenance, security, content drafting, analytics, user support, demand creation and reporting.
• MDH: strategic and technical oversight, validation of all HIV/SRH content and screening tools, verification of reported performance.
• Youth counsellors: respond to support requests within working hours; escalate safeguarding cases.
• Facility focal persons: confirm referral codes in the facility portal at the time of service.
• Institution focal persons & peer educators: promote Uliza, display campus QR codes, support in-institution demand creation.

3. User journey

1. Entry through a campaign link, campus QR code, web or WhatsApp.
2. Language choice (Kiswahili default, English).
3. Consent: plain-language notice on anonymity, limits of the service and aggregate data use. No identifying information is requested at any point.
4. Profile: self-reported age, sex and institution, used only for disaggregated reporting and nearest-facility suggestions. Users under 15 are shown the Child Helpline (116).
5. Use: questions answered from approved content; guided topics; HIV-risk check (MoH HTS screening (adapted) v1.0; MoH GBV screening (adapted) v1.0); service finder.
6. Referral for every user screened at risk, followed by a short satisfaction survey.

4. Urgent & safeguarding cases

• Messages indicating self-harm or suicidal thoughts bypass the AI and return a fixed, MDH-approved safety message with 116 and 112.
• Disclosure of sexual assault or a GBV screen answer of immediate danger or assault within 72 hours triggers urgent guidance: PEP and emergency contraception within 72 hours, care available without prior police report, Police Gender & Children Desk, 116/112, plus an automatic referral.
• A possible HIV exposure within 72 hours is flagged as urgent PEP and referred to a facility offering PEP 24 hours.
• A visible Quick exit control is available on every chat screen.
• Counsellors who identify a child (under 18) at risk of harm follow the national child-protection referral pathway via 116 and the social welfare officer, and inform the MDH focal person within 24 hours without sharing identifying details beyond what the young person has chosen to disclose.

5. Referral & linkage

1. Uliza issues a unique, non-identifying code (format UL-XXXXX) valid for 14 days, naming the service and recommended facility.
2. The young person shows the code at reception. The facility focal person enters it in the facility portal after the service is provided and selects Confirm visit.
3. The platform records the facility and time of linkage and computes days to linkage (≤7 days, 8–14 days, >14 days).
4. Codes not linked after 5 days may trigger one in-chat reminder; no outbound contact is made outside the chat.
5. Referral services available: HIV testing, HIV self-test kit, Condoms, PrEP, PEP (emergency, within 72h), Family planning, STI check & treatment, Mental health support, GBV support.

6. Human support

Users can reach a counsellor by starting a message with MSAADA or SUPPORT. Requests appear in the support inbox; counsellors respond Monday–Saturday 08:00–20:00 EAT, target first response within 2 working hours. Counsellors never ask for names, phone numbers or locations, and use the approved content as their reference.

7. Content governance

• All articles are drafted in both languages from national guidelines with the source recorded, then reviewed and approved by MDH. Only approved articles are available to the AI.
• The AI is instructed to answer only from approved content, never diagnose, never give doses beyond approved content, and to refer when information is not available.
• Quarterly quality checks (minimum 4 per year) sample at least 100 real responses; inaccuracies are logged with corrective actions.
• Screening tools are versioned; changes to MoH tools are applied as data updates after MDH sign-off.

8. Data protection

• Processing is governed by the Personal Data Protection Act, 2022. Data minimisation: no names, phone numbers or IDs stored; WhatsApp numbers are hashed, with the raw number held only in temporary cache for the 24-hour reply window.
• Access is role-based: MDH staff see aggregates and de-identified exports; facility staff see only the referral being confirmed.
• Users may delete all their data at any time from the chat.
• Personal data breaches are reported to MDH within 24 hours and handled per the incident procedure and PDPC requirements.

9. Operations & incidents

• Availability target 99.99% per month. An automated monitor probes the service every minute; two consecutive failures open a P1 incident automatically and recovery closes it, so downtime is measured, not estimated. Maintenance Sunday 02:00–04:00 EAT.
• The database is backed up daily (01:30 EAT) with 30-day retention.
• Every staff sign-in, export, content approval, record change and referral confirmation is written to an audit log reviewed monthly by the super admin.
• Incidents are classified P1–P3 and resolved within 4 hours (P1/P2). Each incident is logged with root cause and corrective action and included in the monthly system performance report.
• If the AI provider is unavailable, the platform automatically serves approved articles so that information access continues.

10. Monitoring & reporting

KPIs (Annex II, Table 1) are computed live on the programme dashboard. Monthly progress updates, monthly system performance reports, quarterly progress and expenditure reports are submitted by the 15th of the month following the reporting period. De-identified exports are available to MDH for verification at any time.