AI Networking Assistant with Semantic Matching

I built a networking platform that stored professional profiles as vector embeddings in Pinecone, enabling semantic search that understands intent ("I need someone who's done B2B SaaS pricing") rather than matching keywords. When a match is found, the system automatically initiates Twilio conference calls between professionals.

Designed the profile embedding pipeline where user attributes (skills, goals, industry, location) are converted into vector representations, enabling similarity search that captures nuance. Built AI tools for extracting profile information from natural conversation, searching across the vector store, and orchestrating multi-party calls via webhooks.

PM Takeaway: The critical UX question in agentic systems: when should the AI act, and when should it ask? Auto-connecting strangers via phone call without explicit confirmation is a terrible experience. This project taught me that the autonomy slider — how much initiative to give the AI — is the most important product design decision in agentic applications.