AgentRec: Personalised LLM-based recommendations

chevron-icon
Back
project-presentation-img
Presentation
Ahan M R
Project Owner

AgentRec: Personalised LLM-based recommendations

Funding Requested

$5,000 USD

Expert Review
Star Filled Image Star Filled Image Star Filled Image Star Filled Image Star Filled Image 0
Community
Star Filled Image Star Filled Image Star Filled Image Star Filled Image Star Filled Image 3.6 (8)

Overview

AgentRec explores the idea of creating a new AI Agent-driven information system tool along with LLMs, to personalise and generated contextually relevant recommendations for a user. In this work, we propose a recommender system based on using LLMs with multi-agent system along with ReAcT framework ("Reason + Then Act") to provide personalized information services supported by user session history and web search. Furthermore, we also propose the development of an early prototype and Low-level architecture document to showcase the exchanging of information between "Items, Agent Recommender as well as the user" to be available as a service on SNET as a personalisation tool in future rounds.

Proposal Description

Our Team

Our team includes experts in Machine Learning and AI Research who have extensive experience working with Large Language Models (LLMs) in the e-commerce industry, providing valuable insights into the specific needs and challenges in these sectors. This knowledge also helps tailor our needs to industry-specific requirements.

View Team

Please explain how this future proposal will help our decentralized AI platform grow and how this ideation phase will contribute to that proposal.

The AgentRec project has the potential to significantly enhance the SingularityNET (SNET) ecosystem by addressing a critical need in the AI and e-commerce sectors: "Contextually aware recommendation systems". AgentRec is a proposed service in the near future on SNET platform, that can achieve the following benefits:

  1. Boosting the variety of services on AI Marketplace: Adding AgentRec to the SNET marketplace will diversify the range of AI services available, making the platform more attractive to potential AGIX users and developers. 
  2. Attracting E-commerce and other small Tech Companies: The demand for advanced RecSys is growing, particularly among e-commerce companies seeking to improve user engagement and conversion rates. AgentRec’s ability to deliver highly personalized recommendations will attract these companies to the SNET platform.

This ideation phase will help design a low-level architecture, along with benchmarking AI agents with LLMs to build an initial prototype of this service.

Clarify what outcomes (if any) will stop you from submitting a complete proposal in the next round.

  1. Performance Limitations: If the LLM-based recommendation system doesn’t perform as expected in terms of accuracy, speed, or scalability during the ideation and prototyping phase, this could indicate some inherent technical issues in the approach and will need longer research plan in place to execute this hypothesis properly. 
  2. Competitive Landscape: New developments or breakthroughs by any competitors in Web2 space could require us to reassess and possibly enhance our solution.

The core problem we are aiming to solve

The core 'problem' AgentRec is focused on is the lack of contextually aware, highly personalized recommendation systems in the current ML space. Most of the commonly used recommendation systems either rely only on user/item or both user and item embeddings to generate predictions for the user, and often fall short in providing more culturally and relevant suggestions due to their inability to understand and interact with users in a context-driven manner. This gap significantly impacts domain focused space such as e-commerce and content-driven companies that rely on effective recommendation systems to enhance customer experience.

Our specific solution to this problem

Enhanced Contextual Understanding using LLM-Based Agent Items:

  • AgentRec uses LLM-based agents that engage in real-time, interactive dialogues with users. This allows the system to capture detailed context and nuanced user preferences that static data alone cannot provide.
  • These agents continuously learn from ongoing interactions, enabling them to understand and adapt to the user's current context and needs dynamically (Adaptive Learning).

Agent-Recommender Collaboration:

  • Agent Items collaborate with Agent Recommenders to refine and enhance the recommendations based on user interactions. This ensures the system remains responsive to changing preferences.
  • AgentRec facilitates collaboration among various Agent Items and Recommenders, allowing them to share insights and user information. This integrated approach enriches the recommendation process with diverse data sources and perspectives.
  • The system’s ability to learn and adapt ensures that recommendations remain relevant and engaging, leading to higher user satisfaction and retention.

Project details

  1. Develop an LLM-based interactive dialogue system to capture user preferences and provide initial recommendations.
    • User Interface: Design a user-friendly interface using Streamlit, Gradio or similar frameworks.
    • LLM Integration: Integrate an LLM such as GPT-4 or RecLlama for natural language processing and understanding.
    • Session Management: Implement session management to track user interactions and maintain context.
  2. Facilitate the exchange of information between Agent Items and Agent Recommenders.
    • Enhanced recommendation accuracy through dynamic collaboration between LLMs and traditional recommender models.
    • Implement coordination algorithms to manage interactions between different agents.

Solution: Multi-Stage Evolution Framework

  • User-Agent Interaction: In the initial stage, users interact with Agent Items through natural language conversations, receiving immediate and contextually relevant recommendations.
  • Agent-Recommender Collaboration: In the second stage, Agent Items collaborate with Agent Recommenders to refine and enhance the recommendations based on user interactions. This ensures the system remains responsive to changing preferences.
  • Agent Collaboration: In the final stage, multiple Agent Items work together to provide comprehensive recommendations, combining expertise from different domains to address complex user needs.

Existing resources

We will leverage several existing technologies and resources:

  • Pre-trained LLMs: Utilize pre-trained LLMs like GPT-4, RecLLM, Transformers4Rec, LLaMA-2, and Mistral-7B to build upon their existing capabilities.
  • Data Repositories: Use publicly available datasets like Wikitext-103 for training and evaluation.
  • Development Tools: Employ development tools and frameworks such as Streamlit for UI, MongoDB for data storage, and Docker and Kubernetes for containerization and orchestration.

Proposal Video

DF Spotlight Day - DFR4 - Ahan M R - AgentRec Personalized LLM based Recommendations

4 June 2024
  • Total Milestones

    3

  • Total Budget

    $5,000 USD

  • Last Updated

    4 Jun 2024

Milestone 1 - System Design and Project Initialisation

Description

Establish the design for foundational architecture and initiate the development of the AgentRec system

Deliverables

This milestone is focused on the low-level and high-level design for integration of LLMs and the development of the interactive dialogue system (ML + Development): (i) System Architecture Design: Design the overall architecture in form of a LLD (low-level document) including how components like the LLM user interface and agents will interact (ii) Interactive Dialogue System (Chat Interface): Develop or utilize a dialogue system that will manage interactions between users and the system for AgentRec

Budget

$2,500 USD

Milestone 2 - LLM Evaluation and Selection for AgentRec

Description

Evaluate various LLMs available in the market to identify the most suitable model for the AgentRec project. This milestone involves setting up the design and initial experiments to test different LLMs with various prompting styles and initial agent prototypes to analyze their effectiveness in understanding user queries and generating accurate recommendations. Literature Review for current approaches and supporting datasets collection will be part of this milestone as well.

Deliverables

(i) Acquisition of e-commerce datasets to test the effectiveness of the Agent-based LLM RecSys models (ii) LLM Evaluation Report: An extensive analysis of tested LLMs detailing their performance metrics strengths and weaknesses in handling personalized recommendation tasks (iii) Documentation explaining the choice of core LLM (open-source vs close-source in terms of latency cost etc) for the AgentRec project supported by data and performance comparisons from the testing phase (iv) Continued development of the prototype of the project with integration of multi-agents with LLMs

Budget

$2,000 USD

Milestone 3 - Final Prototype Development

Description

Deliver a Benchmarking and Evaluation Report along with Initial RecSys prototype with onboarded LLMs

Deliverables

(i) Document that benchmarks the selected LLM against industry standards and evaluates it based on efficiency and accuracy metrics (ii) Ideation Prototype of AgentRec: A basic working prototype of the AgentRec system equipped with the selected AI-agents and LLMs ready to demonstrate personalized recommendation capabilities and gather early feedback.

Budget

$500 USD

Join the Discussion (2)

Sort by

2 Comments
  • 0
    commentator-avatar
    HenriqC
    May 19, 2024 | 3:43 PM

    Thank you for starting from this ideation stage proposal. The deliverables described are exactly what is needed to create a credible full proposal. Such a preference learner would fit well the platform's current needs.

  • 0
    commentator-avatar
    Ubio Obu
    May 16, 2024 | 11:04 PM

    Primarily I am impressed that the proposal fits into the rules of this segment and has a clear-cut vision of what will be deployed to the marketplace eventually. So far it has good prospect and if this personalised LLM works quite well then that will really have lot of use case

Reviews & Rating

Sort by

8 ratings
  • 0
    user-icon
    Ayo OluAyoola
    Jun 10, 2024 | 11:49 AM

    Overall

    4

    • Feasibility 4
    • Viability 4
    • Desirabilty 4
    • Usefulness 4
    AgentRec (Ideation Project)

    AgentRec has the potential to be a powerful personalization tool by combining AI agents, large language models (LLMs), and a user-centric approach. Here are some key points to consider:

    Strengths:

    • Focus on Personalization: Leveraging user data and context to deliver relevant recommendations is a significant advantage.
    • Multi-Agent System with LLMs: This combination offers flexibility and adaptability in understanding user needs and generating recommendations.
    • ReAcT Framework: The "Reason + Then Act" approach provides a structured way for the AI to process information and make informed recommendations.
    • Phased Development: Starting with a prototype and low-level architecture allows for iterative development and testing.

    Points for Further Discussion:

    • Specific Use Cases: Focusing on particular recommendation domains (e.g., e-commerce, news, education) can guide development.
    • Data Privacy and Security: Ensuring user data privacy and security is crucial for building trust with users.
    • Explainability and Transparency: How will AgentRec explain its recommendations to users and foster trust in the AI's decision-making?
    • Evaluation Metrics: Defining how you will measure the success of AgentRec's recommendations (e.g., user engagement, click-through rates) is important.

    Additional Considerations:

    • Scalability: How will AgentRec handle a large number of users and varying recommendation demands?
    • Integration with Existing Systems: Can AgentRec integrate with existing recommendation systems or platforms?

    Overall, AgentRec presents a well-structured approach to personalized information retrieval. By addressing the points for further discussion and considering the additional aspects, the team can develop a valuable tool for various applications.

    Additional Recommendations:

    • Explore existing research on recommender systems and user modeling techniques.
    • Consider potential ethical implications of AI-driven recommendations, such as filter bubbles and bias.

    Have you seen  Ghost AI? Our ideation project just as fantastic as yours. Click https://deepfunding.ai/proposal/persona-ai/ to see what we would like to explore.

  • 0
    user-icon
    Max1524
    Jun 10, 2024 | 2:22 AM

    Overall

    4

    • Feasibility 4
    • Viability 4
    • Desirabilty 4
    • Usefulness 4
    What is the market access strategy?

    Everything is being implemented quite well, I need to know more about the team's thoughts and ideas for the current and future market access strategy combined with expanding the scope of impact. It should be considered an important part of the team's long-term plan.

  • 0
    user-icon
    Tu Nguyen
    May 31, 2024 | 2:15 AM

    Overall

    3

    • Feasibility 3
    • Viability 4
    • Desirabilty 3
    • Usefulness 4
    Personalised LLM-Based Recommendations

    The problem this proposal addresses is the lack of highly personalized, context-aware recommender systems in the current ML space. Their solution was to create a multi-stage evolutionary framework.
    I see that the current project team only has 1 member. They should look for more members to complete this proposal well. In the milestones section, they should identify the start and end times of the milestones.

  • 0
    user-icon
    Rafael_Cardoso
    May 26, 2024 | 9:56 PM

    Overall

    3

    • Feasibility 4
    • Viability 3
    • Desirabilty 3
    • Usefulness 3
    Nice idea but how big is the problem?

    This could be an interesting proposal but to be honest I would need the proposal to focus more on explaining the problem. Why is this problem so big and why people would be willing to pay for this solution to solve that problem.

    That part is really crucial to understanding the potential for the adoption of the solution. and the eventual desirability and impact it could have. 

    It would also be interesting to see why they think they are the best Team to solve this problem and understand why their approach is completely different from other existing solutions

    It would also be interesting to see if they already have any ideas in terms of go-to-market strategy, and if they see any specific companies using this solution, and what would be the strategy to reach these companies.

  • 0
    user-icon
    Gombilla
    May 17, 2024 | 11:28 AM

    Overall

    4

    • Feasibility 4
    • Viability 3
    • Desirabilty 4
    • Usefulness 4
    Strong approach to utilize Advanced AI Techniques

    Incorporating LLMs and a multi-agent system demonstrates a commitment to utilizing advanced AI techniques to deliver innovative solutions, aligning with the marketplace's goal of offering cutting-edge AI services.

    To connect to the aforementioned, by leveraging LLMs and user session history, AgentRec can offer highly personalized and contextually relevant recommendations to users, enhancing their experience within the SingularityNet ecosystem.

  • 0
    user-icon
    CLEMENT
    May 17, 2024 | 11:25 AM

    Overall

    4

    • Feasibility 4
    • Viability 3
    • Desirabilty 4
    • Usefulness 4
    Team shows purpose driven approaches

    Proposing to develop an early prototype and low-level architecture document showcases AgentRec's commitment to delivering tangible results and demonstrates the feasibility of their approach, laying the groundwork for future development and integration within the SingularityNet marketplace. Moreoso, The ReAcT framework, which combines reasoning and action, can enable more intelligent decision-making and adaptive behavior in the recommender system, improving the relevance and effectiveness of recommendations.

  • 0
    user-icon
    mivh1892
    May 17, 2024 | 10:21 AM

    Overall

    3

    • Feasibility 3
    • Viability 3
    • Desirabilty 3
    • Usefulness 4
    impact the e-commerce landscape

    The AgentRec project holds immense potential to positively impact the e-commerce landscape. The project exhibits high feasibility, strong sustainability, significant desirability, and substantial utility. With support from SingularityNET, AgentRec can evolve into a crucial tool for businesses seeking to improve user experience and conversion rates.

    Recommendations:

    • Continue developing the project prototype and gathering user data.
    • Evaluate the recommendation system's performance and make necessary adjustments.
    • Engage with the SingularityNET community to attract support and collaboration.
    • Explore collaboration opportunities with businesses in the e-commerce sector.

  • 0
    user-icon
    Ubio Obu
    May 16, 2024 | 11:07 PM

    Overall

    4

    • Feasibility 4
    • Viability 4
    • Desirabilty 4
    • Usefulness 4
    Can find lot of relevance

    I am seeing how this framework and model if eventually built will find relevance to a lot of other projects built by others, and I can imagine lot of other persons integrating their solution into their apps or projects as well

Summary

Overall Community

3.6

from 8 reviews
  • 5
    0
  • 4
    5
  • 3
    3
  • 2
    0
  • 1
    0

Feasibility

3.8

from 8 reviews

Viability

3.5

from 8 reviews

Desirabilty

3.6

from 8 reviews

Usefulness

3.9

from 8 reviews

Get Involved

Contribute your talents by joining your dream team and project. Visit the job board at Freelance DAO for opportunites today!

View Job Board