Neuroscience-Inspired Memory for AI Agents

Today’s top AI Highlights:

1. Build RAG and AI Agent apps with neuroscience-inspired memory layer

2. RAG and AI agents: Where are AI developers placing their bets

3. All you need is 4x 4090 GPUs to train your own model

4. Get past your Claude limits using your own API key

& so much more!

Read time: 3 mins

AI Tutorials

Ever had your RAG system confidently give completely irrelevant information? Or watched it stubbornly stick to outdated data when better sources were just a web search away? You're not alone. Traditional RAG systems, while powerful, often act like that one friend who never admits when they need to double-check their facts.

In this tutorial, we'll fix that by building a Corrective RAG Agent that implements a multi-stage workflow with document retrieval, relevance assessment, and web search. Using LangGraph's workflow capabilities, we'll create a system that can evaluate its responses, adapt on the fly, and even reach out to the web when its local knowledge falls short. Think of it as RAG with a built-in fact-checker and research assistant.

We'll combine the analytical prowess of Claude 3.5 Sonnet with LangGraph's flexible workflow engine. By the end of this tutorial, you'll have a RAG system that's not just smarter but also more honest about what it knows (and doesn't know).

We share hands-on tutorials like this 2-3 times a week, designed to help you stay ahead in the world of AI. If you're serious about leveling up your AI skills and staying ahead of the curve, subscribe now and be the first to access our latest tutorials.

Don’t forget to share this newsletter on your social channels and tag Unwind AI (X, LinkedIn, Threads, Facebook) to support us!

Latest Developments

Neuroscience-Inspired Memory Layer for LLM Applications 🧠💡

HawkinsDB is a novel approach to AI memory, moving beyond traditional data storage and retrieval. Inspired by neuroscience and Jeff Hawkins’ Thousand Brains Theory, this isn't just another vector database; it's designed for precise, context-aware queries, and manages different types of memory like semantic, episodic, and procedural all in one framework.

To make it easy to work with, specialized packages are built on top of it, HawkinsRAG for RAG and Hawkins Agent Framework for building intelligent agents. It enables developers to move beyond simple similarity searches and build AI applications that truly understand data relationships.

Key Highlights:

1. Context-Aware Queries - HawkinsDB uses reference frames and cortical columns for storing data from multiple perspectives, providing a clear view on how data is connected and decisions are being made. Your AI can handle nuanced queries like "Find kitchen items related to coffee brewing" by understanding objects from multiple angles and connecting related information naturally.

2. Unified Memory - In the framework, semantic, episodic, and procedural memory are not separate modules, but are unified under one system. A customer support AI can now access product specs, past interactions, and troubleshooting guides simultaneously.

3. Tools for RAG and Agents - With HawkinsRAG, you can immediately work on RAG applications using 22+ data sources, and with the Hawkins Agent Framework, you have an ecosystem for quickly building intelligent agents using tools like web search (Tavily), email, and even custom tools. All these are built on top of HawkinsDB, so data handling and memory management are unified in one framework.

4. Memory Control and Asynchronous Design - This system uses SQLite for robust storage, JSON for easy prototyping and gives control on memory retention and entry limits. The framework is designed with modern async/await patterns, for efficient and concurrent operations, which is crucial for scaling agent-based workflows.

5. Start Building - Getting started is straightforward with simple pip install commands for each component. The documentation and quick start examples provide everything you need to begin integrating HawkinsDB, HawkinsRAG, or the Hawkins Agent Framework into your projects.

See Where 3,000+ AI Developers Are Finding Success Now 🤖🚀

As 2024 ends, it's clear that AI agents moved from concept to commonplace this year. If you're aiming to build or enhance your AI applications in 2025, understanding the current state of the ecosystem is essential. We’re drawing insights from Langbase's 2024 State of AI Agents report, directly from over 3,400 developers actively building these systems. This report is a roadmap—it reveals where developers are seeing success, where the bottlenecks are, and what tools and frameworks are gaining momentum.

The report, which is based on 184 billion tokens processed and 786 million API calls, shows where to focus your energy and development efforts. This report offers actionable guidance that can directly impact your development cycle.

1. LLM Selection - OpenAI currently leads (around 76% usage), but Google is rapidly closing the gap (around 59%) with Anthropic also gaining traction (around 47%). Each excels in different areas – OpenAI in tech and marketing, Anthropic in technical tasks, and Google in health and language. Opensource models like Meta's Llama, Mistral, and Cohere, though smaller, are actively growing.

2. Primary Concerns - Scaling complexity is a major roadblock in adopting AI, with developers citing a lack of tooling for monitoring and management along with data privacy concerns. This directly implies the need to focus on robust and customizable solutions for these issues.

3. Development Strategies - Accuracy and customization are key; don't cut corners on testing and optimizing the model's performance. Consider how your workflows can be customized, and remember that cost is the least influential factor when selecting a LLM.

4. Must-Have Infrastructure - Prioritize multi-agent RAG capabilities in your workflows, along with version control and flexible SDKs in your development platform. The data shows that developers highly value these features and are looking for these features in the ecosystem.

5. Top Application Areas - AI Agents, driven by LLMs and often employing RAG, are predominantly being applied in software development (with strong adoption), marketing, IT operations, and text summarization. Customer service, HR, and legal are also emerging as use cases for these technologies.

Quick Bites

A developer documented how to build a local LLM training setup using 4 x NVIDIA 4090 GPUs for around $12,000, showing that consumer graphics cards can effectively train models up to 1 billion parameters. The build, which runs on a custom kernel for optimized GPU communication, includes a Threadripper CPU, 128GB RAM, and handles long training runs with dual 1500W power supplies.

Aitomatic and partners in the AI Alliance released SemiKong, the first open-source LLM specifically trained for semiconductor design and manufacturing. Based on Meta's Llama 3.1 platform, the 70B-parameter model claims to reduce chip development time by up to 30% and dramatically accelerate engineer training through its Domain-Expert Agent system, which can be customized with company-specific technical knowledge.

Tools of the Trade

1. Colada: Chrome extension that uses your Anthropic API key to bypass Claude's daily chat limits while preserving conversation context, making direct requests from your browser. It's a one-time purchase of $9.99.

2. GitDiagram: Open-source tool to generate interactive, clickable diagrams of a public GitHub repository's structure, using information from its file tree and README, rendered in Mermaid.js by Claude 3.5 Sonnet. You can create these by replacing "hub" with "diagram" in any GitHub repository URL.

3. FloTorch: Open-source tool for quickly prototyping and optimizing RAG implementations on AWS, with features for experimentation, performance tuning, and secure deployment. You can configure hyperparameters, manage context, and integrate with LLMs and vector databases for efficient retrieval.

4. Awesome LLM Apps: Build awesome LLM apps with RAG, AI agents, and more to interact with data sources like GitHub, Gmail, PDFs, and YouTube videos, and automate complex work.

Hot Takes

1. Learn to code, biggest roi for me is being a software eng landing in the middle of “smart” coding assistants and knowing how to tell when these models are wrong or right ~
   anton
   
   

2. o1-pro, I changed my mind, after 3 hours of deep discussion on context, plans, and strategy for 2025-2026, I believe it could excel as a Top Manager or even a C-Level. Conversation was challenging on both sides, with insights emerging from the synergy of human and AI reasoning ~
   Ivan Fioravanti

That’s all for today! See you tomorrow with more such AI-filled content.

Don’t forget to share this newsletter on your social channels and tag Unwind AI to support us!

Unwind AI - X | LinkedIn | Threads | Facebook

Awesome LLM Apps | Sponsor Us

PS: We curate this AI newsletter every day for FREE, your support is what keeps us going. If you find value in what you read, share it with at least one, two (or 20) of your friends 😉