Streaming LLM responses to a native iOS client

Fintech & Sports Analytics

Industry: Fintech & Sports Analytics

Platform: iOS

Stack: Swift · Firebase · StoreKit 2 · Combine · SSE

Description: iOS client for an LLM-backed analytics backend with streaming JSON, payment migration, deep links.

---

The engineering surface was a stalled iOS prototype against a Python backend that streamed structured JSON over a long-lived HTTP POST. The buffered approach in the prototype defeated the streaming contract; a custom URLSessionDataDelegate with a brace-depth JSON parser was needed to surface incremental results. Mid-build, the in-app payment system had to migrate to native subscriptions without disrupting existing entitlements, while authentication, push, and deep linking into multiple external targets ran in parallel.

We owned the full iOS client — streaming layer, auth coordination, payment migration, and deep-link routing. The client owned the Python NLP backend and the odds data providers.

Engagement: multi-cycle retainer · single iOS specialist · 2025–2026.

Engineering signal: first prediction card rendered sub-second on stream open; zero-downtime payment-backend migration with existing subscribers preserved.