Ascend OS: Building a Proactive AI Networking Platform
Launched an agentic AI professional networking platform from the ground up: a LangGraph multi-agent system and profile intelligence pipeline that researched 270,000+ professionals.
- 270,000+ professional profiles researched and synthesized by AI
- 50%+ improvement in account creation rate at live events
- LangGraph multi-agent architecture for proactive opportunity discovery
The Situation
Professional networking tools are passive. They store your contacts, show you who’s connected to whom, and wait for you to act. The connections worth making are rarely the ones you’d search for. They’re the ones where what you’re working on happens to need what someone else has, in a way neither of you would have thought to look for.
Ascend OS was built on a different premise: the AI finds those opportunities for you. I was brought in to launch the product and build the team from the ground up.
A New Kind of Agent Architecture
The core of Ascend is a LangGraph-powered multi-agent system built around proactive discovery. Most AI assistants wait for you to ask a question. This one doesn’t.
The architecture has two layers. Background agents run continuously, watching for network changes, surfacing new connections, and scoring opportunities against each user’s goals. When something relevant appears, those agents hand it to the user-facing agent, which delivers it as a notification or message.
Users direct the background agents through conversation. They tell their agent what opportunities to watch for, which relationships matter, what goals to track. The system adjusts. The result behaves like a well-briefed advocate working in the background, not a chatbot waiting to be prompted.
Profiling at Scale
For the matching to mean anything, the system needed to understand people beyond their job title: their goals, how they work, and what they’re actually trying to accomplish.
I designed and built the orchestration system that made this possible. Using Temporal for workflow durability and a LangGraph pipeline for research and synthesis, we discovered and profiled over 270,000 professional contacts, researching each across multiple data sources to build a structured picture of who they are. The “Agent Confidence” score in the product reflects how deeply the system knows each person across dimensions: Goals, Network, Skills, Achievements, Personality, and Culture.
The Event Go-to-Market
Ascend’s target customers are enterprise networks and large organizations building internal connectivity. Live events became a high-leverage entry point.
For each event, agents researched every attendee before arrival, ran full matchmaking across the room, and generated seating assignments based on complementarity. Each attendee received a physical, deck-of-cards-sized packet highlighting specific people they should meet, with a QR code that opened directly to their personal agent in the app.
This flow drove a 50%+ improvement in account creation compared to baseline event conversion. People walking in with a personalized physical artifact connected to a live agent were more likely to sign up and engage.
Outcome
Ascend OS went live with enterprise clients including CEO networks and executive event organizers. The platform has since rebranded as Meshi.