AI
2025
Full-stack Engineer
Nigeria-First AI Ops Copilot
An AI operations copilot for SMBs that reads receipts, categorizes expenses, and supports payment reconciliation in the background.
Problem
The work started here.
Small businesses needed an operations assistant that fits local finance habits and reduces manual bookkeeping around receipts, expenses, and payments.
Research
- Studied how SMB owners track expenses through receipts, transfers, and informal records.
- Identified where categorization and reconciliation could remove repetitive admin work.
- Shaped the product around practical operations rather than generic enterprise assumptions.
Planning
- Defined receipt capture, expense categorization, and reconciliation as core workflows.
- Planned AI assistance around reviewable suggestions instead of opaque automation.
- Built the product as a focused web app that could be tested quickly with users.
Architecture Highlights
- Next.js and React power the product interface and workflow shell.
- AI workflow logic turns receipt and payment inputs into structured suggestions.
- Vercel deployment supports fast iteration and public demos.
System Design
- Receipt records move from capture to categorization to review.
- Reconciliation flows compare payments and expense records before confirmation.
- User-facing explanations keep suggested categories understandable.
Challenges
- The assistant needed to feel local and practical, not like a generic finance demo.
- AI suggestions had to be useful without hiding uncertainty from users.
- Expense workflows needed to stay quick for busy business owners.
Solutions
- Centered the UX on receipts, categories, and reconciliation tasks SMBs already recognize.
- Kept AI output reviewable so users can correct or confirm suggestions.
- Reduced the interface to the decisions users need to make next.
Performance
- Delivered a public AI copilot prototype for SMB operations.
- Improved bookkeeping flow by connecting receipt intelligence to reconciliation tasks.
- Kept the product lightweight enough for fast iteration.
Tech Stack
Next.js
React
TypeScript
AI Workflows
Vercel
Code Snippet
const insight = await categorizeReceipt({
receiptText,
merchant,
paymentReference,
});Gallery & Animations
The final product uses subtle hover tilt, reveal choreography, and responsive image transitions to make the case study feel alive without turning the page into a heavy demo reel.
Lessons Learned
- AI products land better when they match the user's real operating environment.
- Automation should make financial work easier to verify, not harder to understand.
- Local context is a product requirement, not a marketing detail.