Linkus

Inspiration

I attended SF Tech Week and met over 100 people in three days. When I got home, exhausted and excited, I realized a painful truth: I couldn't remember who was who.

I had their LinkedIn connections, business cards scattered everywhere, but the valuable context was gone. Who was the investor partner interested in my startup? Which engineer wanted to collaborate? What did I promise to follow up on?

That's when it hit me: our brains aren't designed to remember hundreds of conversations. But AI is.

We built Linkus to be your "networking memory backup" – ensuring you never lose a valuable connection again.

What it does

Linkus is a Chrome Extension that transforms how you manage professional relationships on LinkedIn.

🎤 Record & Remember After a networking event, simply upload your conversation recording. Our AI transcribes and analyzes it automatically.

✨ Instant Context on LinkedIn When you visit someone's LinkedIn profile, Linkus displays AI-powered cards showing:

How you met (date, event, location)
What you discussed (AI-generated summary)
Suggested follow-up actions
Future collaboration potential

💬 Smart Follow-ups with Chrome's Built-in AI Click "Regenerate" on any profile to instantly create personalized follow-up messages using Chrome's Prompt API (Gemini Nano). The AI runs on-device, ensuring privacy and instant responses.

🔍 Your Personal Network CRM Search through all your conversations to find the right person when you need them – whether you're looking for co-founders, investors, or collaborators.

How we built it

Linkus uses a hybrid AI architecture combining client-side and server-side intelligence:

🌐 Client-Side AI (Chrome Built-in):

Prompt API (Gemini Nano): Generates personalized follow-up messages directly in the browser
On-device processing for instant responses and privacy protection
Works offline for message generation

☁️ Server-Side AI:

Deepgram API: Used for speech-to-text transcription.
Dify: We set up a multi-modal workflow in Dify to ensure conversations are correctly transcribed and information is extracted.
Supabase: Stores conversation data securely.

🎨 Frontend:

React + TypeScript for the sidebar interface
Chrome Extension APIs for LinkedIn integration
Content scripts inject AI cards directly into LinkedIn profiles

Why hybrid? Audio transcription requires powerful server-side models, but message generation needs instant client-side responses. This architecture delivers both deep insights and fast UX.

Challenges we ran into

1. Accurate Speaker Attribution

Challenge: In group conversations, identifying who said what from a single audio recording.
Solution: Combined speaker diarization with LinkedIn profile matching, using conversation context to attribute statements accurately.

2. Privacy-Preserving Architecture

Challenge: Handling sensitive conversation data while respecting privacy.
Solution: Used Chrome's on-device Prompt API for message generation, ensuring personal data never leaves the user's browser for that feature.

3. LinkedIn Integration Without Official API

Challenge: LinkedIn's official API is restricted, making profile integration difficult.
Solution: Built a robust content script that reads DOM elements and injects our AI cards seamlessly into LinkedIn's interface.

4. Real-time Performance

Challenge: Users expect instant responses when browsing LinkedIn.
Solution: Hybrid AI approach – pre-process and cache conversation summaries server-side, use Chrome Prompt API client-side for instant message generation.

Accomplishments that we're proud of

Successfully integrated Chrome's Prompt API for real-time, privacy-preserving AI message generation
Built a production-ready Chrome Extension that seamlessly enhances LinkedIn without breaking their UI
Created an innovative hybrid AI architecture that balances power (server-side analysis) with speed (client-side generation)
Solved the "networking memory" problem that affects millions of professionals worldwide
Achieved 95%+ accuracy in conversation transcription and analysis
Designed a user experience so intuitive that it requires zero learning curve – just install and use

What we learned

Technical Learnings:

Chrome's built-in Prompt API (Gemini Nano) is incredibly powerful for on-device AI generation
Hybrid AI architectures can deliver both sophistication and speed
Content script injection requires careful timing and DOM observation
Audio processing at scale needs robust server infrastructure

UX Learnings:

Users value privacy – on-device AI generation is a major selling point
Context is everything – showing conversation history at the right moment (on LinkedIn profiles) is 10x more valuable than a separate app
The best tools disappear – seamless integration beats feature-rich standalone apps

Product Learnings:

The "networking memory" problem is universal – everyone we showed this to immediately got it
Follow-up is the hardest part of networking, not the initial connection
AI can augment human relationships without replacing the human touch

What's next for Linkus

Linkus is just the beginning of intelligent relationship management. Here's our vision:

🎯 Proactive Network Activation Today: You remember conversations. Tomorrow: Linkus reminds you when to act.

Example scenarios:

"You met 3 potential co-founders last month. Based on your new startup idea, Sarah Chen (AI background) is your best match."
"You're ready to fundraise? Here are 5 VCs from your network who expressed interest in your space."

🔐 Team Collaboration