WORK

African merchants lose sales at the seam between fiat and crypto. Accepting both means two providers, two settlement flows, and two reconciliation headaches.

One checkout that weaves fiat and crypto into a single thread. Customers pay with their bank (card, transfer, USSD) or any wallet on any chain. The merchant settles in stable dollars or local currency without seeing the seam. Public demo and developer docs.

Live at paywithweave.com in early access, onboarding merchants.

Fintech ops teams answer the same questions all day: is this provider degraded, why did this transaction fail, what do we retry. The data exists; querying it takes an engineer.

A command center for fintech and payments teams: AI answers over live operational data, provider health advisories, proactive alerting, and approval-gated recovery workflows. Informed by the lead engineer's production experience building AI operations tooling in fintech.

In development as an upcoming Dewdrop Labs product.

Nigerian homes run on a mix of unstable mains, inverters, and solar. When supply drops, everything competes for the same watts and the inverter dies early.

An ESP32-based residential controller with RS485/Modbus RTU inverter integration, TRIAC load switching, and zero-cross detection. It sheds low-priority loads automatically as available power changes, designed for Nigerian mains and hybrid solar systems.

Related research co-authored and published at IECEC 2026, University of Nigeria, Nsukka. Hardware in active development.

PCB designers leave their tool every time they need help. Context is lost; cloud AI is slow or unavailable on weak connections.

An AI assistant embedded in the EasyEDA Pro design environment, running on a local Ollama backend. LLM help arrives inside the hardware design workflow, offline-friendly by design.

Static defenses give attackers unlimited time to study a target. The system never changes; the attacker only improves.

Adaptive moving target defense using eBPF and reinforcement learning: the system reshapes its own attack surface faster than it can be mapped.

Student-focused businesses need real e-commerce (inventory, payments, orders) without the cost of a custom build or the ceiling of a generic storefront.

An e-commerce platform built by Dewdrop Labs, originally delivered under subscription licensing. FLOF Mart has since bought out the platform and owns it outright.

Live in production. Dewdrop Labs now serves as FLOF Mart's engineering partner, building out the rest of their product.

Smart mirrors and AR beauty try-on tools ship either as closed, cloud-dependent hardware or as disconnected demos. Gesture control, voice assistants, and AR overlays rarely work together, and rarely run without streaming video to someone else’s servers.

A privacy-first AI mirror that fuses a MagicMirror² display frontend with a Python computer-vision backend: MediaPipe hand tracking for gestural control, real-time AR lipstick overlay triggered by touch or voice, and a local voice assistant (Ollama running Llama 3.2) for wake-word commands, weather, search, and music. All AI processing runs locally; no video is persisted to the cloud.

Working prototype (v9) with gesture, AR, and voice pipelines running end-to-end on a single machine.

Developers already have the context they need sitting on disk: code, docs, notes. Querying it means breaking flow to open a browser tab, and cloud AI tools mean sending that code somewhere else.

A native macOS AI agent: a Swift UI with a global hotkey sits above every app, backed by a local Python brain running a quantized Qwen3 model on Apple Silicon via MLX. It indexes local folders for retrieval-augmented search, and an agentic loop can find, read, and propose edits to files, reviewable in a Draft Mode before anything is applied. Nothing leaves the machine.

Public v0.1 release with a macOS installer, alongside build-from-source instructions.