Projects

Two live products, plus the supporting artefacts built along the way.

AI-Powered KYC Address Verification

For Nigerian Fintechs

Claude VisionDeterministic RulesRAGEval HarnessCBN Regulations

$ eval-runner --cases 22
 
v1.0  Precision: 72.7%  Recall: 88.9%
      F1: 80.0%  Accuracy: 72.7%
 
v1.1  Precision: 90.0%  Recall: 100.0%
      F1: 94.7%  Accuracy: 95.5%
 
Δ  +22.8pp accuracy · 0 false approvals

The biggest failures weren't technical — they were Nigerian-specific. The system rejected customers named "Hajiya Zainab Danladi" because "Hajiya" wasn't in the title-stripping list. Evaluation-driven improvement isn't optional for a product built for Nigerian users.

Live Demo Case Study →

Agentic Cross-Border Trade Compliance Advisor

Nigeria → China Corridor

Agentic AILangChainRAGCBN RegulationsStreaming UI

$ corridor-agent --transaction TX-001
 
[✓] Document extraction    COMPLETE
[✓] Supplier verification  VERIFIED
[⚠] Sanctions screening    FATF flag
[✓] Regulatory retrieval   4 citations
 
VERDICT: COMPLIANCE_READY
Refs: CBN/TED/FEM/PUB/FPC/01/003
      NCS Circular 01/2022

A state-machine agent, not a chatbot. The compliance engine verifies, screens, and cites — then routes uncertain cases to human review. Every verdict is traceable to a specific CBN or NCS circular. "Automate the workflow, not the accountability."

Live Demo Case Study →

Also built along the way

AI Feature Teardown Report

Carbon · Mono · Copilot Money

Analysed AI/ML systems in three fintech products: credit scoring, transaction enrichment, personalised categorisation. Five cross-cutting patterns across all three.

CarbonMonoCopilot MoneyML Systems

View Report →

LLM Eval Pipeline

LLM-as-Judge · Anthropic API · Python

Automated evaluation of KYC reasoning quality across 5 cases using Claude as judge. Source grounding scored 4.4/5 — finding directly informed the governance policy.

LLM-as-JudgePythonAnthropic API

View Report →

Fintech RAG Prototype

LangChain · ChromaDB · CBN Regulations

Built a working RAG system that answers questions about Nigerian fintech regulations using CBN and Consumer Protection documents. 108 chunks, semantic search.

LangChainChromaDBCBN Regulations

View Report →

Prompt Library v1.1

Prompt Engineering · Fintech

10 production-ready prompts for a Nigerian lending chatbot. Versioned with changelog, temperature recommendations, and compliance-specific instructions per use case.

Prompt EngineeringLendNaijaFintech

View Report →

LendNaija KYC Scenario

Applied AI Concepts

Applied tokenisation, embeddings, temperature, and context window concepts to a realistic Nigerian MSME lending scenario. Phase 1 portfolio deliverable.

TokenisationEmbeddingsMSME Lending