Skip to content
Back to portfolio

Personal AI Product

AI Networking Platform

A personal AI product that turns recorded networking conversations into prep, analytics, and feedback.

December 2025 – January 2026

Problem

Users needed AI-generated networking support, including conversation preparation, analytics, and feedback from recorded audio.

My contribution

Developed a Python FastAPI backend and a React Native frontend backed by MongoDB, integrated GPT-4o and Whisper speech-to-text through a streaming audio pipeline exposed via REST APIs, extracting talking points, action items, and tone feedback.

Simplified architecture

Simplified: a streaming audio pipeline transcribes recorded conversations with Whisper and uses GPT-4o to extract talking points, action items, and tone feedback, served through a FastAPI backend, MongoDB, and a React Native app.

Outcomes

  • Validated with a 42-user beta.
  • 78 percent D7 retention.
  • 61 percent activation.
  • 95 percent extraction accuracy using token-level F1 on a 300-sample held-out set.
  • Under 4-second p95 end-to-end latency.
  • Approximately $0.006 per audio minute.
  • Error analysis identified overlapping speech and domain jargon as primary failure modes.

Technologies

PythonFastAPIReact NativeMongoDBGPT-4oWhisperREST APIsStreaming audio