Projects

Designing and building a white-label, multi-tenant event app platform from scratch as solo founder. Brands (festivals, conferences, churches, cultural organisations) launch branded iOS/Android apps backed by a shared admin dashboard, API, push notification system, Eventbrite integration, and analytics — without bespoke engineering. Currently in pilot with the first real-world client (ThisChord), with live marketing, admin, and API services in production.

• –Designed a multi-tenant org → brand → event hierarchy with tenant-safe data access, shared admin surface, and per-brand theming and layout variants so one platform change improves every branded app
• –Shipped an installation-based push notification system with 4 categories, audience targeting (everyone / account holders / event followers), scheduled delivery, and open-rate metrics
• –Built an Eventbrite OAuth + bulk import + webhook sync pipeline so brands can ingest entire event catalogues with price and availability kept in sync via a background scheduler
• –Implemented a full attendee account lifecycle — sign-up, social login, provider link/unlink, password change, and 30-day soft-delete with email cancellation — enforced across every auth path (including OAuth) with 410 termination on scheduled accounts
• –Running active R&D on FriendFinder, a BLE/mesh proximity discovery feature for 70k-person events, with a deterministic simulation harness and 9 architecture, privacy, and feasibility specs
• –Running the full product stack end-to-end as solo founder — native apps, admin, API, infra (Hetzner + Coolify + GHCR), observability (GlitchTip, Uptime Kuma), pricing, and go-to-market

Built a digital language-preservation platform for Kimbundu by transforming a historical scanned Kimbundu–Portuguese dictionary into a structured, auditable lexical corpus. Designed a multi-stage pipeline covering PDF page extraction, column segmentation, OCR capture, deterministic parsing, corpus reconstruction, conservative LLM auditing, and editorial merge workflows. Produced a final merged corpus of 10,679 entries with provenance and review tracking, then published a website-ready public dataset powering kimbundu.org.

• –Built a multi-stage OCR → corpus reconstruction pipeline for a historical dictionary spanning hundreds of scanned pages
• –Produced a final merged editorial corpus of 10,679 entries with provenance, cleanup metadata, and review workflows
• –Published a public dictionary dataset and website experience to support Kimbundu language preservation