Cold Start Killer

Submission Document — Cold Start Killer

Submitted by: Tammy (Solo Participation)

Hackathon: Progress x GitNation Hackathon 2026

Generated with some help from Claude AI.

The Problem

Every developer knows the feeling. You're standing in a room of 500 brilliant engineers, designers, and builders, and you have no idea who to talk to or what to say. Conference networking is broken. Attendees scroll through massive schedules, miss the people most relevant to them, and leave without making the connections that could actually change their careers.

This isn't an information problem. It's an anxiety problem. The tools that exist today (Brella, Grip, Swapcard) give you lists. Lists don't tell you what to say when you walk up to a stranger.

The Solution

Cold Start Killer is an AI-powered conference networking co-pilot that visualises the entire social graph of an event and tells you exactly who to talk to — and what to say when you do.

How it works:

1. You enter your name and select your technical interests

2. The app renders a live force-directed network graph of all attendees, with connection strength visualised as edge weight and colour

3. You click any attendee node to see your shared interests, an interest overlap radar chart, relevant sessions to attend together, and an AI-generated conversation opener personalised to both of your profiles

4. As new attendees join, their nodes animate into the graph in real time

How Kendo UI is Used

Cold Start Killer uses Kendo UI React components as core interactive elements throughout the experience:

• Kendo UI Chart (Radar) — displays interest overlap between the user and any selected attendee, making compatibility immediately visual

• Kendo UI Grid — displays shared and recommended conference sessions with time and room data

• Kendo UI Inputs — powers the profile tag selection interface

• Kendo UI Layout — structures the sidebar panel and card components

• Kendo UI Buttons — styled action buttons throughout the interface

Kendo UI's theming system is used to create a cohesive dark-mode design language consistent across all components.

Technical Stack

AI Integration

The AI layer uses Groq's Llama 3 model to generate personalised, context-aware conversation openers in real time. Each opener is generated using both attendees' names, roles, companies, and shared interests — producing openers that feel human and specific rather than generic.

The AI prompt is engineered to:

• Address the target person by first name

• Reference something specific about their role or company

• Stay under 2 sentences for immediate usability

• Avoid corporate language and generic phrases

What Makes It Different?

Existing tools like Brella and Grip show you ranked lists of people you might want to meet. Cold Start Killer shows you the entire social topology of the room at a glance, and then solves the hardest part — the moment you actually have to walk up to someone and open your mouth.

The force-directed graph makes patterns immediately visible: clusters of interest, bridge nodes who connect different communities, and your own position in the social landscape of the event.

Real World Path

In production, Cold Start Killer would integrate with conference registration APIs to populate the graph with real attendee data. Attendees opt in during registration, select their interests, and the graph is live from the moment the doors open.

The simulate attendee feature, which animates new nodes into the graph in real time, demonstrates exactly how this would feel as registrations come in.

What's next?

• Real conference API integration (JSNation, React Summit attendee data)

• QR code check-in that triggers your node appearing on others' graphs

• Post-conference follow-up: which connections did you actually make?

• Organiser dashboard showing community clusters and topic gaps