LAI #91: Reinforcement Learning, Knowledge Graphs, and Modular AI Agents

Author(s): Towards AI Editorial Team

Originally published on Towards AI.

Good morning, AI enthusiasts!

This week’s issue highlights how reinforcement learning and modular architectures are reshaping AI systems. We feature new research applying RL to sequential market basket decisions, showing how Q-learning can optimize for long-term value rather than one-off predictions. You’ll also find a knowledge graph fusion framework that integrates LLMs for more accurate reasoning, a hands-on tutorial for building stock investment agents with MCP, and a guide to multi-touch attribution models, from Shapley values to LSTMs, revealing how different methods shift which marketing channels get credit.

Let’s get into it!

What’s AI Weekly

This week, I have something you won’t want to miss: we’ve launched a referral program where you can earn real rewards. Share with friends, and you could win our physical book or get free access to any of our courses. Plus, top referrers get an invite to our affiliate program, where you can start earning up to $70 for every course you recommend.

Sign up here or find more details here!

— Louis-François Bouchard, Towards AI Co-founder & Head of Community

Free for O’Reilly Members: 10-Hour LLM Fundamentals Bootcamp

If you’re an O’Reilly member, you can get free access to our 10-hour LLM Fundamentals course: a one-day, language-agnostic Bootcamp built for software professionals.

You’ll learn exactly when to Prompt, RAG, Fine-Tune, or Deploy Agents, and walk away with the practical knowledge needed to work with LLMs in real projects.

This five-part, bingeable video series covers:

• Foundational AI knowledge and using LLMs
• Building on top of LLMs
• Evaluating RAG and LLM pipelines
• AI workflows and agents with real case studies
• Guardrails, advanced techniques, optimizations, and monitoring

🎯 Who it’s for: Software professionals ready to add LLM skills to their toolkit.

💡 What you get: A clear roadmap to move from basics to production-ready AI systems.

👉 Start learning now on O’Reilly

Learn AI Together Community Section!

Featured Community post from the Discord

Superuser666_sigil has created an MCP-server fuzzer. It is A CLI-based fuzzing tool for MCP servers using multiple transport protocols, with support for both tool argument fuzzing and protocol type fuzzing. Check it out on GitHub and support a fellow community member. You can also test it on your MCP server and share your feedback in the thread!

AI poll of the week!

Nearly 6 in 10 people openly cite ChatGPT when they use it, whether in school, work, or projects. That’s a big shift in just a couple of years: AI isn’t just a hidden assistant anymore, it’s something people are increasingly comfortable acknowledging. What do you think: in 5 years, will citing AI tools be as normal as citing Google, or will it still feel like something we need to justify? Tell me in the thread!

Collaboration Opportunities

The Learn AI Together Discord community is flooding with collaboration opportunities. If you are excited to dive into applied AI, want a study partner, or even want to find a partner for your passion project, join the collaboration channel! Keep an eye on this section, too — we share cool opportunities every week!

1. Cpnk75m is working on an AR + AI app for the Devpost Hackathon and is looking for someone who can help with picking the right ML models, some quick hacks to get it running, AR objects/modals for overlays, and solid testing. If you want to join and build this project into something bigger, connect in the thread!

2. Efficientnet_99825 is looking for someone to write a research paper, pick a dataset from Kaggle and work on it, build and finetune LLMs, and work on the stock market, quant analysis, and time series. If this sounds interesting, connect with him in the thread!

3. Silentsentinel6943 is looking for a partner to help him grow his GitHub repo, MARM-Systems. He needs help with scaling. If you can help, reach out in the thread!

Meme of the week!

Meme shared by rucha8062

TAI Curated Section

Article of the week

Beyond Associations: Reinforcement Learning for Sequential Market Basket Decisions By Shenggang Li

Moving beyond static product associations, this research applies reinforcement learning (RL) to optimize sequential product recommendations. The approach models shopping as a decision-making process, using customer clustering, contextual bandits, and Q-learning to create policies aimed at maximizing business value, such as margin. These new policies are then evaluated against traditional market basket analysis baselines using off-policy evaluation methods like SNIPS and Doubly Robust to safely estimate their impact from historical data. The results suggest that the Q-learning models provide a significant performance lift by optimizing for long-term value throughout a customer’s entire shopping session.

Our must-read articles

1. Graph Fusion + KGC + LLM Agents = Powerful AI Reasoning By Gao Dalie (高達烈)

Addressing the limitations of traditional knowledge graph construction, which often produces disconnected and inaccurate sub-graphs, the Graph Fusion framework is introduced as a more integrated solution. It employs a three-step process: seed entity extraction using BERTopic, candidate triplet generation via a large language model, and a global knowledge fusion module. This final step is crucial for merging similar entities, resolving conflicts, and identifying new relationships across various texts. The result is a more unified and precise knowledge graph, offering a more efficient method for integrating scientific knowledge from unstructured text.

2. MCP 101 Tutorial: Build Your Own Modular AI Agent for Stock Investment Insights By Lorentz Yeung

This article provides a tutorial on creating a modular AI agent for stock investment analysis using the Multi-Component Protocol (MCP). The system architecture relies on a large language model to route user queries to specialized servers that handle tasks like mathematical calculations and stock data retrieval. It details the setup process and provides the necessary code for this base framework. It also discusses how this foundation can be expanded with additional modules, such as sentiment analysis or predictive models, to build a more sophisticated financial analysis tool.

3. Multi-Touch Attribution — A Quick And Practical Guide By Jonty Haberfield

Assigning credit for customer conversions across multiple marketing touchpoints is a complex challenge. This piece examines several Multi-Touch Attribution (MTA) models, comparing a basic last-touch baseline with more advanced techniques like Shapley values, Markov Chains, and an LSTM neural network. By applying these methods to the same dataset, it was demonstrated that the resulting channel attributions vary significantly. The analysis highlights how the choice of model directly impacts which channels are valued, offering guidance on selecting a method based on data complexity and whether the sequence of interactions is important.

4. Reinforcement Learning for Agentic AI: Optimizing Decision Making By Samvardhan Singh

To address decision-making in unpredictable environments, this analysis shows how Reinforcement Learning (RL) improves agentic AI. It reviews RL fundamentals, such as Markov Decision Processes, and their application in knowledge graph navigation. The central idea is the integration of RL into LangGraph, which turns workflow nodes into adaptive decision-makers. A hands-on tutorial for an AI tutor and a logistics routing optimization case study demonstrate this approach. The logistics example, in particular, details how an agent adapts to real-time traffic data, offering a practical template for developing intelligent, responsive systems.

If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI