W
Wispr Flow Tech Stack
Voice-to-text transcription with speech model personalization and low-latency inference
Wispr Flow is a voice transcription product shipping across Mac, iPhone, Windows, and Android. The stack reveals a consumer-first, multi-platform architecture (TypeScript, React, Python, Swift, Kotlin) paired with heavy investment in speech ML (personalization, LLM inference, reinforcement learning optimization). Pain points cluster around scaling: personalizing speech models, reducing inference latency, and handling high request volumes — indicating the company is moving from single-user accuracy toward billions-of-users infrastructure.
Tech Stack 16 technologies
Core StackTypeScript React Python Swift Kotlin Zendesk Intercom Google Ads AppsFlyer iOS Android Bing Ads Apple Search Ads Google App Campaigns Google Analytics 4 PostHog
AdoptingAndroid
What Wispr Flow Is Building
◆Challenges
- Scaling engineering team
- Personalizing speech models at scale
- Scaling personalization of speech models
- Scaling llm inference latency
- Breaking asr performance boundary
- Scaling voice interface
- Low-latency inference infrastructure
- Handling high request volume
- Low-latency inference for millions of requests
- Scaling to a billion users
▲Active Projects
- Prototyping new features of voice interface
- Llm inference infrastructure
- Speech model personalization
- Voice interface development
- Voice interface for billions of users
- Rl-based personalization of llms
- Creator partnership program
- Content strategy execution
- Growth channel optimization
- Building first capable voice interface
Hiring Activity
Minimal20 roles · 2 in 30d
Department
Seniority
Notable leadership hires: Recruiting Lead
Company intelligence
