LinkedIn Curator AI: Your Knowledge Base Builder

Architecture Diagram
LinkedInAI Assistant Outcome

Intro

Linkedin is a professional networking and social sharing app and provides professionals a platform to connect and share information. Linkedin’s Creator functionality allows individuals to create specialised content, courses and short-long form content based on their area of expertise. Today, the platform - although really engaging and informational, is crowded with professional opinions, learning resources and news.

Users engaging with such content find it difficult to resist the doom scroll with articles, posts and webinars emerging every second. Being intentional with reading long form content,video and even topics and conversations seems impossible.

In order to save time, while also managing “Fear of Missing Out”, users save posts to read later. They continue to engage with posts, but forgo the opportunity to learn and retain information from the posts, often forgetting to go back and read again. Users of such social platforms remain inundated with information, and need a way to keep engaging, yet also find a sustainable way to save and retain information and insights

How we built it

After teaming up, we realised that we come from diverse backgrounds in tech, product management and sales even. This helped us articulate the problem we faced, and drafted a first level problem statement that eventually got built into a Product Requirements Document. In parallel, we used ADK quickstarts on our local machines and also versions on Firebase, to test how LLM agents worked and how agents in general worked. A lot of the code written, was written with heavy duty prompting using Google Vertex AI and Open AI along with failed attempts with Replit as the front end.

Challenges

We then realized that a Chrome extension to capture LinkedIn posts would be the best option. We met regularly over the week to discuss the status of our individual attempts and group efforts too. Finally, after we got the agents to work, we documented our learnings and tweaked and added an additional feature to add comments via an AI agent too!

Finding project partners from a large WhatsApp group and making sense to each other is one of the greatest achievements we made. From 3 different continents, and diverse backgrounds, technology united our efforts.
Successfully delivering a hackathon especially starting from a place of knowing nothing about AI development, to creating running agents and setting up environments on Google Cloud, it felt magical and mind we say, extremely rewarding!

From a code perspective, we ran into a problem invoking the outcome of the LLM agents. To invoke the LLM capability, we need to define a base agent and build a custom function to experience the result of the LLM capability. for eg,invoke(),call () had challenge on the same.

What's next

Going forward we want to continue to build further with Google ADK, and see how we can productionize the code. Further we want to build an end-to-end use case for a B2B user working in a corporate workplace along with other workplace technology and integrate it there. We are looking forward to making other learners learn effectively and in an inexpensive way!