Petr Rusanov

I took MealSnap from idea to a released iOS app with live users, using AI to help people build intuitive eating habits with less effort. The iOS app turns food photos into editable entries and estimates, recalculates after corrections, and provides personalized weekly and monthly reviews. I developed the Go/MongoDB backend and rebuilt the client in SwiftUI with coding agents in 2026. I also designed an agent skill for UI automation in iOS Simulator and on physical devices, providing a verification loop with screenshot and video evidence. The AI workflow uses structured results, validation and repair, bounded retries, timeouts, circuit breakers, and Gemini/OpenAI failover. I evaluate prompt changes against a defined scenario corpus with two independent AI judges checking usefulness, tone, and factual grounding. In the October 2026 snapshot, the app had approximately 500 monthly active devices and approximately 7,500 sessions in 30 days.

The old frontend stack of an internal vehicle tool had become difficult to maintain, and development had nearly stopped. I directed coding agents to rebuild it in React while preserving its structure and behavior. I defined acceptance criteria and used browser comparisons of old and new pages, screenshots, and user interactions to find discrepancies. The agents added E2E tests where the old tool had none. A first ready version took about two weeks, followed by a two-week colleague pilot and fixes. The team adopted the replacement, and further development became faster.

I built a personal AI assistant to help organize family life: briefings, activities, weather and weekend planning over Telegram. Grocery preparation through Picnic is a supporting workflow. I developed it in TypeScript/Node.js on pi.dev. Its memory combines SQLite, sqlite-vec, FTS5 and local Ollama embeddings for hybrid vector/keyword retrieval, relevance/recency/importance ranking, and keyword fallback when embeddings are unavailable. I designed scoped tools, permitted-user checks, secret filtering, a read-only container filesystem and allowlisted outbound access. Grocery preparation stops before order placement, keeping checkout with a person. The project uses cloud models alongside local memory and embeddings.

I built the REST API backend and companion iOS app for a paperless ticketing system used by concrete suppliers and their clients. The system records electronic-ticket statuses and shows trucks on a live map. Clients can access tickets, track deliveries, place orders and pay bills.

I led a cross-functional team developing an English-learning iOS app built around real-world content and a user-contributed dictionary. The work included a redesigned iPad app, frequent design reviews, and a public beta with 150 testers.

A smart-home app that communicates with a hub and cloud services using Protobuf. Its interface includes custom controls for adjusting light brightness and RGB colors, so users can control connected devices.