Kwesi Owusu

B Online

• Migrated full e-commerce platform from WooCommerce to custom Next.js 14 + Laravel 11 architecture. 184 database models. 5 state machines. 159 async background jobs. 16+ Docker containers.
• Built GraphQL API covering products, orders, inventory, and fulfillment. Integrated DEAR ERP via webhooks. Multi-currency payments (ZAR, GBP, USD) with PeachPayments. Elasticsearch-powered search with faceted filtering.
• Built internal sales platform for leads, customers, purchases, and order fulfillment (pick, pack, ship). Started as frontend engineer building the customer-facing site. Now full-stack across all services.

B Online

• Built multi-agent AI sales assistant: 3-layer system with NL-to-SQL and hybrid vector + SQL search. Role-based data access. Used daily by the sales team. Also built AI product photo generation using Google Gemini with smart prompt engineering per product category.
• Built AI document validation (GPT-5 vision) for ID/passport extraction. Automated product descriptions (GPT-4o). Automated brand/model/style/material extraction from product names. All running as background jobs in production.
• Built product embedding pipeline for semantic search across the catalog. Incremental sync from AWS S3 via Celery. 9 AI features total across the Luxity platform.

B Online

• Built production RAG system for ASSA. Ingests website content and PDF library via WordPress REST API. Hybrid search using HNSW vector + GIN full-text indexes on pgvector.
• 6-layer instruction system with agentic tool loops (re-searches if first attempt insufficient). AI-powered conversation flagging detects frustrated users. Source citations link to original PDFs and web pages.
• Built eval suite using Agno's AccuracyEval (LLM-judged, 7+ score threshold). Production Docker Compose with resource limits (512MB/0.5 CPU per service). PgBouncer connection pooling. Sentry monitoring.

B Online

• Built Duolingo-style employee training platform from scratch. 26,000+ lines of TypeScript across 3 apps (API, learner app, admin dashboard). Organizations upload documents. GPT-4 generates courses, levels, and questions.
• Gamification system: points (+10 correct, -5 incorrect), badges, daily streaks, leaderboards. Daily cron detects rank drops and sends targeted email notifications. Multi-tenant with organization-scoped access.
• Document processing pipeline handles PDF, DOCX, and TXT. BullMQ async job queue with Bull Board monitoring. Sentry error tracking. Role-based auth via Supabase.

Personal project

• I do copywriting on the side and write with data — real customer quotes, pain points ranked by frequency, objections with counter-evidence. The research was taking 20-40 hours per project. So I built a pipeline that does it in 10 minutes.
• 5 AI agents run in parallel via Inngest. Each searches 100+ sources (Reddit, G2, Trustpilot, Hacker News). Output feeds into 13 parallel assembly groups. Final result: 23-section research brief with frequency analysis.
• Real-time progress via SSE. Stripe credit system. Playwright PDF generation. PostgreSQL with Drizzle ORM. Live at brevvi.ai.