Projects

Email campaign platform with a paying client — full pipeline from block editor to inbox.

• Block editor → email-safe HTML renderer with a pixel-accurate preview before send.
• SES fan-out send pipeline: QStash-queued, idempotent workers, suppression-loop handling for bounces and complaints.
• 200,000+ emails sent to date, with sending capacity well beyond that.

One command replaces your entire dev infrastructure setup — Artemis provisions persistent, production-grade database clusters locally so your environment matches prod from day one.

• Kubernetes-native operator: spins up PostgreSQL, Redis, and MongoDB clusters from declarative manifests, managing persistent volume claims, pod lifecycle, and stable connection endpoints across restarts.
• Full-stack web dashboard (Next.js + TypeScript) with live database introspection via native drivers — real-time health indicators, connection strings ready for direct .env consumption.
• Ships as a single distributable binary via esbuild, published to npm.

On-device Android scam detection system with real-time ML inference and system-level call interception.

• Custom Android launcher with system-level call interception and real-time on-device ML inference.
• No cloud dependency, sub-200ms detection latency.
• Hybrid React Native + Kotlin architecture using Android's native telephony APIs.
• Foreground service persistence across device lifecycle.

Passive, local, zero-API memory for AI coding agents — with a 3D desktop app to visualise it.

• Watches Claude Code transcripts and extracts meaningful moments (decisions, file changes, error fixes) — the agent never participates in the write path.
• On-device semantic embeddings via candle (all-MiniLM-L6-v2, 384-dim) — no cloud, no API keys, ever.
• Hybrid search: BM25 keyword + cosine semantic, fused with Reciprocal Rank Fusion.
• 3D memory graph (react-force-graph-3d + three.js) — orbit, zoom, and fly through a project's decision history.
• Secret redaction before storage; dual persistence to SQLite + human-readable Markdown.
• Ships an MCP server so any MCP-capable agent can recall memory automatically.

AI-powered event planning platform with LLM-driven workflow generation.

• LLM-driven workflow generation converting event inputs into structured planning timelines.
• ~70% reduction in manual planning time.
• ~99.5% uptime via Firebase + Cloudflare.