Report

Production Prompt Library v1.1

Prompt Engineering · LendNaija · Nigerian Fintech

Overview

Built a library of 10 production-ready prompts for LendNaija — a Nigerian MSME lending AI assistant. Each prompt is versioned, temperature-specified, and designed for a specific compliance or customer-facing use case. The library demonstrates how prompt engineering is a product decision, not a technical afterthought.

The Principle

Every line in a system prompt is a product decision — tone, boundaries, compliance rules, output format, and user experience are all defined through the prompt. The PM owns the prompt. Engineering owns the plumbing.

The 10 Prompts

$ prompt-library --version 1.1 --count 10
 
Transaction Categorisation      Few-shot + confidence scoring
  → Classify ₦ transactions
Loan Eligibility Assessment     Role-play + few-shot + CoT
  → Credit decisions
Credit Score Explanation        Role-play + few-shot
  → Explain decisions to users
Fraud Detection Flagging        Zero-shot + structured constraints
  → Flag suspicious activity
KYC Document Verification       Zero-shot + checklist + CoT
  → Verify identity docs
Customer Complaint Routing      Few-shot + confidence threshold
  → Route to right team
Loan Repayment Reminder         Role-play + few-shot (staged)
  → Reduce defaults
Risk Assessment Summary         Role-play + structured + debug mode
  → Internal risk tool
Regulatory Compliance Check     Role-play + CoT + CBN references
  → CBN compliance
Financial Literacy Tip          Zero-shot + constraints + anchor
  → Customer education

V1.0 → V1.1 Improvements

1. Added temperature recommendations to every prompt
2. Added confidence scoring to classification prompts
3. Added few-shot edge cases for Nigerian name patterns
4. Added debug mode toggle to risk assessment prompt
5. Added repetition control to financial literacy tips
6. Added quantified thresholds to fraud detection

Key PM Insight

Temperature is a product configuration, not a technical default. Different features within the same product need different settings:

Credit decisions — 0.0

compliance-critical, deterministic

Fraud detection — 0.0

safety-critical

Repayment reminders — 0.3

slight variation, not robotic

Financial literacy tips — 0.8

creative content, variety desired

Nigerian Regulatory Context in Prompts

The Regulatory Compliance Check prompt explicitly references:

CBN AML/CFT/CPF 2022
Nigeria Data Protection Act 2023
FCCPC Digital Lending Guidelines

This domain specificity is what separates a production prompt from a generic template.