lvl100Sign in

Girl Effect

Request for Proposal: Agentic AI System Partner Discovery, Design, Build and Handover for WAZZII

  • Kenya
  • Full-time
  • Posted

Skills

  • Cybersecurity
  • Microsoft Azure
  • Procurement

Job description

Procurement Summary

Item                                   Details

RFP title                          Agentic AI System Partner for WAZZII Kenya

Procurement method      Open competitive tender

Eligible applicants          Qualified technology firms or consortiums with agentic AI, responsible AI and production-engineering capability

Location                         Kenya, with remote delivery permitted and in-person Nairobi workshops as required

Contract period             Up to 12 months: three-month discovery phase and an optional nine-month build phase

Contract gate                Phase 2 will proceed only after Girl Effect’s written Go or Re-scope approval

Clarification deadline    20th November 2026

Consolidated FAQ       25th November 2026

Proposal deadline      2nd December 2026

 

Who We Are

Girl Effect is an international non-profit that builds media and technology girls choose, use and trust. Across Africa and South Asia, our youth brands — including Wazzii and Tuki in Kenya, Big Sis in South Africa, Chhaa Jaa and Bol Behen in India, and Yegna in Ethiopia — reach millions of adolescent girls and young women (AGYW) with content and digital tools that support the decisions shaping their lives, from sexual and reproductive health (SRH) to wellbeing, education and work. Our approach is girl-centred by design: girls co-create our products, our evidence teams measure what actually changes for them, and our safeguarding architecture treats their safety as a design constraint, never an afterthought.

Girl Effect has spent a decade building conversational products for young people, and several years deliberately and publicly charting what responsible AI for girls looks like. Our published Ethical AI Guidelines, our AI/ML Vision and our GenAI Whitepaper set out the standards all our technology partners are held to; they are annexed to this TOR and should be read before responding.

Background

WAZZII is Girl Effect’s WhatsApp-based chatbot for adolescent girls and young women in Kenya. It provides confidential sexual and reproductive health information, signposting and access to human help-desk support in English and Sheng.

Human operators using the RapidPro help desk remain central to the service.

Girl Effect intends to evolve WAZZII into a trusted service-navigation platform that can help users:

•Find vetted youth-friendly service providers.
•Prepare for a service-provider visit.
•Access services through approved referral mechanisms.
•Receive appropriate follow-up and continuation support.

To achieve this goal, we envisage the use of an agentic AI solution, consisting of multiple AI-powered features serving specific use cases which work under the instruction of an orchestrator. This solution will proactively support a girl’s end-to-end journey, offering guidance, follow-up, referral coordination and continuation support within strict safety and consent boundaries. This TOR selects the partner for that exploration and, if approved, that build.

Existing Technology Environment

Girl Effect has an existing Azure-hosted intelligence environment that manages interactions between its digital products and large language models.

The selected partner will be required to build on the existing codebase and work closely with Girl Effect’s DevOps, product and programme teams to deploy, monitor and manage all approved components.

Responsible AI Context

The assignment must align with Girl Effect’s:

•Ethical AI Guidelines.
•Guidelines on Ethical AI-Powered Social and Behaviour Change Chatbots.
•AI and Machine Learning Vision.
•GenAI Whitepaper.
•Safeguarding, data-protection and technology policies.

These documents will form part of the procurement and contractual requirements.

Objective and Intended Outcomes

Girl Effect seeks an experienced partner to investigate, design, prototype and, subject to a formal approval gate, build an agentic AI layer that supports a user’s end-to-end WAZZII journey within strict consent, safeguarding and human-escalation boundaries.

The engagement should deliver:

•A justified, secure and vendor-conscious architecture that works with Girl Effect’s existing Azure environment and codebase.
•Safe and consent-based agentic capabilities for approved WAZZII use cases.
•Clear human oversight, escalation and intervention mechanisms.
•An evaluation framework covering safety, accuracy, usefulness, efficiency and behavioural quality.
•Production-grade observability, including decision logs, traces, dashboards, alerts and audit records.
•Transparent prototype, pilot and scale cost models.
•Complete code, configuration, documentation and knowledge transfer, enabling Girl Effect to operate, audit and extend the solution independently.

What You will Do

The partner will investigate, prototype and — post-gate — build the agentic layer for the end-to-end girl journey, working closely with Girl Effect’s product, safeguarding, evidence and service design teams. As part of the project, an Evaluation Framework will need to be designed which is able to measure the efficacy of the different AI-powered features and of the system as a whole, with key metrics available to the program team through a dashboard. 

Phase 1 — Discovery (months 1–3). Key activities include:

•Architecture investigation — assess candidate architectures for an agentic layer on WAZZII, with trade-offs across capability, safety, controllability, efficiency, observability and vendor/model dependencies — options with trade-offs, not a single foregone conclusion
•Proof-of-concept — build a thin agentic slice against real (sandboxed) WAZZII flows — for example proactive follow-up and referral coordination — demonstrating feasibility, not polish
•Safety case — define how agentic behaviour stays inside Girl Effect’s escalation triggers, consent rules and safeguarding protocol: failure modes, mitigations, and the assurance approach, reviewed by Safeguarding
•Evaluation framework alignment — define how the system will be measured against Girl Effect’s evaluation frameworks — behavioural quality, safety, accuracy and efficiency — including the evaluation suites and benchmarks to be built in Phase 2
•Observability design — specify the instrumentation of agent behaviour: decision and action logging, tracing across multi-step journeys, dashboards and alerting, and the audit trail Girl Effect requires
•Efficiency and cost modelling — model running costs at pilot and at scale, aligned to Girl Effect’s evaluation frameworks
•Phase 2 build plan — scope, sprint plan, team, interfaces with the platform workstream, and acceptance criteria

Phase 2 — Build (months 4–12, contingent on gate approval). Key activities include:

•Iterative build — of the accepted architecture in two-week sprints, against the acceptance criteria set at the gate
•Evaluation suite implementation — the evaluation framework from discovery made live: automated evaluations and benchmarks run per release, with results reported to Girl Effect
•Observability implementation — logging, tracing, dashboards and audit trails live in production; agent behaviour reviewable by Girl Effect at any time
•Integration — with the platform workstream via interfaces defined at inception, each party naming an integration lead
•Safety testing — as sprint deliverables, with no capability touching girls going live until the safety case is satisfied
•Documentation and handover — such that Girl Effect can operate, audit and extend the system independently

Deliverables

Stage

Deliverables

Onboarding

Write up of understanding of GE’s existing AI technology and its application to the use case

Discovery

Architecture options paper; working proof-of-concept; safety case (safeguarding-reviewed); evaluation framework and benchmark definitions; observability specification; efficiency and cost model; Phase 2 build plan

Gate

Girl Effect’s written Go/No-Go/Re-scope decision. No payment attaches to the gate itself

Build (post-gate)

Sprint releases against acceptance criteria; evaluation suites live with per-release results; observability live (logging, tracing, dashboards, audit trail); safety tests passing per release

Handover

Production-ready agentic capability operating within the safety case; full documentation; handover enabling independent operation, audit and extension by Girl Effect

Girl Effect will accept a deliverable only where it:

•

Is complete.

•

Meets the agreed acceptance criteria.

•

Is supported by the required evidence.

•

Has no unresolved critical or high-severity defects.

Correcting incomplete or non-compliant work will not constitute a change in scope or attract additional fees.

Management and Working Arrangements

•The engagement will be managed by Girl Effect’s Product and Delivery Lead, supported by designated safeguarding, programme, evidence, DevOps and integration leads.
•The partner must appoint an accountable engagement lead, solution architect, AI/ML lead, safety lead and engineering lead.
•Work will be managed through Girl Effect’s agreed project-management platform.
•The partner will participate in weekly delivery meetings, two-week sprint reviews and monthly programme demonstrations.
•The partner will maintain the work plan, backlog, decision log, risk register, dependency log, architecture records, evaluation results and cost tracker.
•Material changes to the architecture, model, data, scope, budget, personnel, security controls or sub-processors require Girl Effect’s prior written approval.
•Named key personnel may not be replaced without Girl Effect’s approval and an equivalent or more qualified replacement.

Who You Are

The successful bidder should demonstrate:

•At least three years’ relevant experience delivering production AI or advanced data systems.
•Recent experience with agentic or tool-using AI systems.
•Strong safety-engineering and responsible AI experience, preferably within health, youth, public-service, financial or other sensitive contexts.
•Demonstrated experience in AI evaluation, red teaming, tracing, observability and auditability.
•Experience integrating with existing cloud architectures, APIs, messaging platforms and DevOps environments, preferably Microsoft Azure.
•Strong cybersecurity, privacy and data-protection capability.
•Experience designing technology for young people or vulnerable populations, which will be an advantage.
•Capacity to work with Kenya-based stakeholders and complete the full discovery-to-handover assignment.

Proposal Submission Requirements

Bidders must submit separate technical and financial proposals (Max 10 pages).

Technical Proposal

It should include:

•Understanding of the assignment, intended outcomes, risks, and dependencies.
•Proposed discovery methodology.
•Architecture investigation and proof-of-concept approach.
•Responsible AI, safeguarding, privacy, security, evaluation and observability approach.
•Indicative Phase 2 architecture and implementation methodology.
•Detailed Phase 1 workplan and indicative 12-month plan.
•Proposed team, roles, location, level of effort, availability and CVs
•Three relevant case studies, including at least one deployed AI system and one sensitive-domain assignment.
•Proposed models, platforms, tools, open-source components, licences and subprocessors.
•Approach to intellectual property, portability, documentation and handover.
•Key risks, assumptions, exclusions and information required from Girl Effect.
•Three contactable client references.

Financ

Browse all jobs by role, city or company