Prompt Engineer

Design, author, and maintain production-grade prompts for AI voice bots deployed across outbound debt collections campaigns and inbound collections IVR in English and Swahili, optimizing prompt structure and tuning them on platforms like Gemini 2.5 Flash and OpenAI GPT-4o.

1. Design, author, and maintain production-grade prompts for AI voice bots deployed across outbound debt collections campaigns and inbound collections IVR in English and Swahili.
2. Optimise prompt structure using Markdown formatting principles (headers, delimiters, instruction hierarchy) to maximise model comprehension and minimise token overhead and call latency.
3. Tune and test prompts on Gemini 2.5 Flash (Google AI Studio / Vertex AI) and OpenAI GPT-4o, benchmarking outputs against latency, coherence, and collections compliance targets.
4. Evaluate how prompt length, token count, instruction complexity, and model parameters (temperature, max_tokens, streaming) affect inference latency and per-call cost.
5. Implement and iterate on system prompts, few-shot examples, persona definitions, objection-handling branches, promise-to-pay capture logic, and fallback scripts for edge-case scenarios.
6. Conduct structured QA audits on sampled collections call recordings, scoring against defined parameters (script adherence, tone, promise-to-pay handling, loan disclosure compliance).
7. Maintain a versioned prompt change log with A/B test configurations and measurable outcome data; produce regular performance summaries for the Call Center Manager.
8. Coordinate with the DC IT Technical Support Officer on API configurations, Webitel IVR routing, SIP trunk settings, and system-level constraints that affect prompt deployment.
9. Collaborate with collections strategy teams to translate DPD bucket cadence rules and product-specific scripts (Boda, Tuk-Tuk, smartphone loans) into accurate bot logic.
10. Liaise with AI vendors on ASR quality, voice synthesis settings, and integration constraints relevant to Kenya collections operations.
11. Monitor per-call cost metrics and propose prompt changes that improve recovery rates and ROI without compromising quality or compliance.
12. Any other duties that may be assigned.

1. Minimum 1–2 years of hands-on experience in prompt engineering, conversational-AI / chatbot design, NLP, or a closely related role. Candidates with strong, demonstrable project work and shorter formal tenure may be considered.
2. Demonstrated experience designing and deploying prompts or dialogue flows for production LLM or voice-bot systems — not solely experimental or academic work.
3. Track record of tuning and evaluating prompts against measurable KPIs and running structured A/B tests to improve real-world outcomes.
4. Prior exposure to contact-centre, telephony, or IVR environments, ideally in collections, customer service, or financial services.
5. Experience in or alongside debt collections, lending, fintech, BPO, or banking operations is a strong advantage.
6. Experience collaborating with cross-functional teams (IT, strategy, operations) and managing third-party AI / voice-vendor relationships is desirable.
7. Experience producing English- and Swahili-language conversational content for a Kenyan or East African customer base is highly desirable.
8. Bachelor’s degree in Computer Science, Software Engineering, Computational Linguistics, Data Science, or a related field. Equivalent, demonstrable hands-on experience in prompt engineering or conversational AI will be considered in lieu of a degree.
9. A reviewable portfolio of production prompt work — prompt libraries, system prompts, evaluation harnesses, or published repositories — that can be walked through at interview.
10. Certification in Google Cloud (Vertex AI / Generative AI), OpenAI, or an equivalent LLM / conversational-AI program is an added advantage.
11. Hands-on command of core prompt-engineering techniques: system and role prompting, few-shot and chain-of-thought design, delimiter and Markdown structuring, output schema and constraint control, and guardrail / fallback design.
12. Working proficiency with at least one major LLM platform — Google Gemini (AI Studio / Vertex AI) and/or OpenAI GPT-4o — including parameter tuning (temperature, max_tokens, top-p, streaming) and a practical grasp of tokenization and context-window economics.
13. Understanding of voice-AI pipelines — automatic speech recognition (ASR) and text-to-speech (TTS) — and how transcription accuracy, latency, and barge-in behaviour affect live call performance.
14. Familiarity with IVR / telephony concepts: Webitel or comparable contact-centre platforms, SIP trunking, call routing, and API-based integrations.
15. Ability to read and reason about JSON API request/response payloads and to work in API consoles (e.g., Postman). Basic scripting in Python or JavaScript for prompt testing and batch evaluation is an an advantage.
16. Competence in designing and interpreting A/B tests and quantitative metrics — latency, containment, promise-to-pay rate, recovery rate, and cost-per-call.
17. Full professional fluency in English and Swahili, with the ability to author, review, and QA bot dialogue and call scripts in both languages.
18. Awareness of debt-collections conduct standards and Kenyan consumer-protection and data-privacy obligations, including the Data Protection Act, 2019 and CBK consumer-protection expectations.
19. Strong analytical and problem-solving mindset with meticulous attention to detail.
20. Comfortable with ambiguous requirements and rapid, data-driven iteration.
21. Clear written and verbal communicator who can translate technical findings for non-technical stakeholders.
22. Self-driven and organised, able to maintain a versioned change log across multiple concurrent campaigns.