IntelliFeed AI: Your Personalized Content Curation Co-Pilot with Deep Dive Summaries

IntelliFeed AI is an intelligent assistant designed to combat information overload and deliver a highly personalized content experience. Initially, it learns and adapts to individual user preferences by ingesting diverse data sources (e.g., ChatGPT history, YouTube history, newsletters) and employs a CrewAI multi-agent system to provide personalized recommendations of videos, articles, and other online content.

Building upon these initial recommendations, IntelliFeed AI introduces a LangGraph-powered deep dive summarization feature. When a user selects a specific YouTube link from the recommendations, a dedicated LangGraph workflow is initiated. This workflow includes a specialized Summarizer Agent that leverages the user’s preference memory and the context of all previously recommended content to generate a nuanced summary of the selected video. A subsequent Verify Agent then ensures the quality and relevance of this summary before it’s presented to the user. This hybrid approach combines broad content discovery with focused, context-aware summarization.

Users interact with IntelliFeed AI through a modern web interface built with Next.js, while the backend operations, including data storage for user preferences and feedback, are powered by a Neon serverless Postgres database. The system’s continuous learning capability, driven by user feedback on both initial recommendations and deep summaries, ensures that all aspects of content delivery remain relevant and aligned with evolving user interests.

We have successfully implemented a full-stack application for IntelliFeed AI, featuring:

User Interface (Frontend):

• Built using Next.js, providing a responsive and interactive experience.
• Allows input of initial preference indicators.
• Displays curated content recommendations (articles and videos) from the CrewAI system.
• Allows users to select a specific YouTube link for a “deep dive” summary.
• Presents the LangGraph-generated and verified summary for the selected video.
• Features feedback mechanisms for both initial recommendations and deep summaries.

Backend Services & Database:

• Leveraging Neon (serverless Postgres) for data storage (user profiles, preference memory, feedback).
• Handles API requests and orchestrates both CrewAI and LangGraph agent executions.

AI Core - Part 1: CrewAI for Initial Recommendations:

• PreferenceLearning Agent: (Role: Preference Researcher) Collects and maintains user interests.
  Task (setup_interests): Establishes a structured user profile.
• ContentDiscovery Agent: (Role: Content Discovery Specialist) Searches for relevant online articles and YouTube videos.
  Tasks: discover_articles, discover_videos.
• SummaryGeneration Agent (CrewAI): (Role: Content Aggregator) Compiles discovered content into a summary report.
  Tasks: aggregate_filter_content, generate_summary (for initial recommendations list).

AI Core - Part 2: LangGraph for Deep Dive YouTube Summarization:

• Triggered when a user selects a YouTube link from the CrewAI recommendations.
• Summarizer Agent (LangGraph):
  Goal: To generate a comprehensive and context-aware summary of a specific YouTube video link.
  Capabilities: Accesses user preference memory (from Neon DB), the full list of initially recommended links/articles (for broader context), and the target YouTube video content.
  Process: Analyzes the selected YouTube video in light of the user’s known interests and the thematic landscape of other recommended content to produce a highly relevant summary.
• Verify Agent (LangGraph):
  Goal: To validate the accuracy, relevance, and quality of the summary generated by the Summarizer Agent.
  Process: Checks the summary against predefined criteria or potentially against the source video content to ensure it’s a good and faithful representation.
• LangGraph State & Flow: Manages the state (e.g., selected link, raw summary, verified summary) and orchestrates the interaction between the Summarizer and Verify agents.

Feedback Loop Implementation:

• Users provide feedback via the Next.js UI on both initial recommendations and the deep dive summaries.
• Feedback is stored in the Neon database and used by the PreferenceLearning agent (CrewAI) and potentially informs the context for the LangGraph agents in future interactions.

Demonstration:

1. User inputs preferences via Next.js UI.
2. CrewAI agents generate and display a list of personalized content recommendations.
3. User selects a specific YouTube link from this list for a deep dive.
4. The LangGraph workflow is triggered:

Summarizer Agent (LangGraph) generates a summary using user preferences and overall context.  
Verify Agent (LangGraph) checks the summary.  

5. The verified deep summary is displayed to the user in the UI.
6. User provides feedback on recommendations and/or the deep summary.

(b) Theme: Practical Agents - How is it on theme?

IntelliFeed AI strongly embodies the “Practical Agents” theme through its dual-layered multi-agent architecture:

• CrewAI System for Broad Personalization: The initial set of agents (PreferenceLearning, ContentDiscovery, CrewAI’s SummaryGeneration) practically handles the broad task of sifting through information and providing initial personalized recommendations.
• LangGraph System for Focused Tasks: The LangGraph implementation showcases practical agents for more specific, stateful, and potentially iterative tasks like deep summarization and verification of a single piece of content. The Summarizer and Verify agents in LangGraph have clear, practical roles in this focused workflow.
• Collaborative Workflows: Both systems demonstrate collaboration. CrewAI agents pass information sequentially. LangGraph agents operate within a defined graph, potentially with cycles and conditional logic, to refine the output (summary).
• Specialized Roles & Goal-Oriented Behavior: Each agent, whether in CrewAI or LangGraph, has a defined role, goal, and utilizes specific tools or knowledge (e.g., user preference memory for the LangGraph Summarizer).

This hybrid approach, using different agent frameworks for different scales of tasks, is a practical solution to a complex problem.

(c) Impact: How is this impactful in the real world if it were real?

The impact remains significant and is further enhanced:

• Combating Information Overload: Still a core benefit.
• Enhanced Learning and Discovery: The deep dive summaries provide an even quicker way to grasp the essence of a video, making learning more efficient.
• Personalized Professional Development: Professionals can quickly get context-aware summaries of relevant talks or presentations.
• Improved Content Vetting: The Verify Agent adds a layer of quality control to the summaries, increasing trust.
• Deeper Understanding with Context: The LangGraph Summarizer, by using broader preference and content context, can provide summaries that are not just accurate to the video but also highlight aspects most relevant to the specific user.

(d) Creative/Novel/Innovative: How is this novel, creative, and something that hasn’t been seen before?

IntelliFeed AI’s novelty and innovation is now even more pronounced:

• Hybrid Agent Architecture (CrewAI + LangGraph):
  Novelty: The combination of CrewAI for broad, initial content discovery and recommendation, followed by a LangGraph implementation for a stateful, deep-dive analysis and summarization of selected content, is a sophisticated and novel approach. It leverages the strengths of both frameworks: CrewAI for orchestrating a team of specialized agents for a broader task, and LangGraph for more complex, graph-based execution flows for focused tasks.
• Context-Aware Deep Summarization:
  Innovation: The LangGraph Summarizer doesn’t just summarize a video in isolation. Its access to user preference memory and the context of all other recommended articles/links allows it to generate summaries that are uniquely tailored and highlight aspects of the video most pertinent to the user’s overall interest profile. This goes beyond generic summarization.
• Integrated Verification Step: The inclusion of a Verify Agent within the LangGraph flow to explicitly check the summary quality before user presentation is a creative step towards building more reliable and trustworthy AI assistants.
• Holistic Preference Ingestion & Continuous Learning: These existing strengths are now applied to both layers of the system.

This multi-layered agent system, where different types of agent orchestrations are used for different granularities of tasks, all while maintaining a cohesive user experience and learning loop, is highly creative and innovative.

3. SYSTEM WORKFLOW DIAGRAM (TEXT-BASED)

1. User Interaction & Initial Preference Input:

[User] --- Interacts with --> [Next.js UI]  
[Next.js UI] --- Submits Initial Preferences / Data Sources --> [Backend API]  

2. Profile & Preference Memory Storage:

[Backend API] --- Sends Data --> [Neon DB]  
[Neon DB] --- Stores/Updates User Profile & Preference Memory --> (Data Ready)  

3. CrewAI Agent Workflow for Initial Recommendations:

(Input: User Profile Data from Neon DB)  
[PreferenceLearning Agent (CrewAI)]  
    --> Task: setup_interests  
    ---- (Output: Structured User Profile)  
[ContentDiscovery Agent (CrewAI)]  
    (Input: Structured User Profile)  
    --> Task: discover_articles  
    ---- (Output: Articles List)  
    --> Task: discover_videos  
    ---- (Output: Videos List)  
[SummaryGeneration Agent (CrewAI)]  
    (Input: Articles List, Videos List)  
    --> Task: aggregate_filter_content  
    ---- (Output: Refined Content Collection)  
    --> Task: generate_summary_report  
    ---- (Output: Formatted Recommendation List)  

4. Display Initial Recommendations:

[SummaryGeneration Agent (CrewAI)] --- Formatted Recommendation List --> [Backend API]  
[Backend API] --- Sends Recommendations --> [Next.js UI]  
[Next.js UI] --- Displays Curated Recommendation List --> [User]  

5. User Selects YouTube Link for Deep Dive:

[User] --- Selects YouTube Link via --> [Next.js UI]  
[Next.js UI] --- Sends Selected Link --> [Backend API]  

6. LangGraph Workflow for Deep Dive YouTube Summarization:

(Input: Selected Link, User Preferences & All Recommendations Context from Neon DB/Memory)  
[Summarizer Agent (LangGraph)]  
    --> Generates Summary of Selected YouTube Link  
    ---- (Output: Generated Summary)  
[Verify Agent (LangGraph)]  
    (Input: Generated Summary)  
    --> Validates/Refines Summary  
    ---- (Output: Verified/Refined Deep Summary)  

7. Display Deep Summary:

[Verify Agent (LangGraph)] --- Verified Deep Summary --> [Backend API]  
[Backend API] --- Sends Deep Summary --> [Next.js UI]  
[Next.js UI] --- Displays Verified Deep Summary --> [User]  

8. Feedback Loop:

[User] --- Provides Feedback (on Recommendations and/or Deep Summary) via --> [Next.js UI]  
[Next.js UI] --- Submits Feedback --> [Backend API]  
[Backend API] --- Sends Feedback Data --> [Neon DB]  
[Neon DB] --- Stores Feedback --> (Feedback Recorded)  
(Feedback Data from Neon DB) --- Informs next cycle for --> [PreferenceLearning Agent (CrewAI)]  

Explanation of the Text-Based Workflow Diagram:

• The workflow is broken down into numbered stages.
• [Component Name (Framework if specific)] indicates a system component.
• –> or — Action —> indicates data flow or process initiation.
• Indentation shows sub-processes or tasks.
• (Input: …) and (Output: …) clarify data.

NA