Private AI Chat

Vanish

What began as a hackathon project built in roughly five days using Cursor and Expo evolved into a production iOS and Android subscription private AI app. The project gained strong internal traction and was ultimately launched as Vanish, a private AI chat app, in June 2025. At the time of writing, Vanish has been well received, with a 4.9 rating on the App Store.

Background

By the end of 2024, it was clear that search was shifting toward AI based products ChatGPT, Claude, Perplexity, and the like. But as adoption grew, so did concerns about privacy. People were handing their questions and context to these services without strong guarantees about how that data would be used. I saw an opening: a private AI chat that could serve both the web and mobile, without forcing our existing Startpage users to use AI within existing search. Early AI search had disappointed many with hallucinations; RAG-style approaches were changing that. Perplexity, for instance, combined real Bing results with AI and cited sources, making answers more trustworthy. We validated the idea with a small iOS audience, and when an internal “AI week” came up, I pitched GhostChat. Using Cursor and Expo, I had a basic app on internal TestFlight within days.

I adapted the Startpage app icon in Figma to create the GhostChat icon. With only light React experience, I used Cursor to direct the build, navigate changes, and shape the product.

My Contributions

• Originated, prototyped and pitched the initial concept for the app, which later evolved and was rebranded as Vanish
• Partnered closely with design to shape the core product vision, UX direction, and privacy-first experience
• Defined product strategy and requirements, and led cross-functional development
• Launched the free version of Vanish in June 2025 to validate market demand and user behavior
• Orchestrated iOS and Android builds, managing end-to-end App Store and Google Play delivery
• Evaluated, tested, and integrated multiple AI providers — including Perplexity (Sonar), Claude, OpenAI, and Gemini — to ensure flexibility and performance
• Introduced geo-based trending prompts to reduce the “blank page” problem
• Initiated rollout and promotion to the existing Startpage audience, leveraging built-in distribution to accelerate adoption

App in stores

When I exited the company in January 2026, the app maintained an exceptional 4.9 rating across both the App Store and Google Play. An uncommon achievement for an AI product in a crowded market. This was accomplished with a lean team consisting of one frontend and one backend developer.

What did I learn?

Shipping while running the engine

Building a new AI product alongside a live core business demanded disciplined scope, shared resources, and intentional trade-offs. Speed of learning mattered more than polish.

Models change, architecture endures

We avoided early self-hosting and instead orchestrated across OpenAI, Claude, Gemini, and Perplexity. This abstraction layer gave us resilience as providers shifted and prevented lock-in.

Privacy must be structural

Trust was earned through system design, data handling, retention, and provider controls. Not through promises or positioning alone.

Additional Screenshots

These screenshots were taken on iOS Jan 2026.

Note: The information shared here is based on my direct contributions and publicly available details. Additional aspects of this project are intentionally omitted to respect company confidentiality.