Ultra‑Low Latency Local‑First Interview Copilot

Description and use

WishPilot is a local‑first interview copilot delivering sub‑second speech intelligence using Electron, React 19, and Groq Whisper Turbo, with zero telemetry and local key storage.

Problem

Conventional conversational AI assistants suffer from 8–10 s latency and require uploading private audio and credentials to cloud servers, which is unacceptable for live interviews and raises privacy concerns.

Outcome

WishPilot achieves sub‑second speech intelligence on a desktop, eliminates privacy risks by keeping all data local, and provides a real‑time interview assistant with customizable answer refinement.

Workflow

• Capture microphone audio via AudioWorkletNode
• Downsample to 16 kHz PCM and run VAD
• Send audio chunk to Groq Whisper Large v3 Turbo over HTTPS
• Receive STT and synthesize context (resume, domain, vision)
• Stream LLM response through Unified AI Streaming Engine (BYOK)
• Display answer in floating HUD with WDA_EXCLUDEFROMCAPTURE
• Offer quick‑action pills for answer refinement

Evidence

The spoken question is transcribed with technical vocabulary accuracy in under 250 milliseconds.

Source

WishPilot has zero central backend servers. No telemetry pings, no user tracking, and no database holding interview transcripts.

Source