Engineering logs &
architecture plans.

The Challenge: Biljakt required a system capable of aggregating vehicle listings from Sweden's major auto marketplaces (Blocket, Wayke) in real time, executing natural language semantic matching, and cross-referencing registration records to protect buyers from fraud.

The Solution: We engineered a high-throughput scraping engine powered by Puppeteer clusters. The backend uses OpenAI's GPT models to interpret complex search briefs, returning the top 3 best-matched options alongside parsed historical safety metadata.

The Challenge: Context switching friction when using web AI chats and the need to extract text locally from uncopyable contents like images, videos, and PDFs, while keeping conversation history completely private.

The Solution: We built a Google Chrome Extension using Manifest V3 that runs a persistent side panel React app. It integrates Tesseract.js for local OCR, Web Speech API for real-time transcription, and IndexedDB via Dexie.js for secure, private local logs.

The Challenge: Implementing a storage-efficient rolling 30-second video buffer capture system in React Native (Expo Bare Workflow) that merges dynamic 3-second segments and corrects camera rotation offsets without using deprecated libraries.

The Solution: We developed a custom Expo Native Module using Swift (AVFoundation) and Kotlin (MediaMuxer/MediaExtractor). JS handles the rolling buffer queue and triggers the JSI module to perform hardware-accelerated merging and rotation correction on native system threads.

The Challenge: Integrating a native Apple Watch (watchOS) app target into an Expo React Native workspace, facilitating real-time bi-directional messaging (WCSession) and preventing Expo prebuild from wiping target configurations.

The Solution: We built a native communication module linking JavaScript and watchOS via WCSession. A custom Config Plugin coordinates Xcode project targets during EAS prebuild, and a Git-isolation strategy safeguards targets from getting overwritten.