Drishtant Ghosh (Drix10)
drix10blogs
GitHubLinkedIn
AI Companies and Ventures••7 min read•1205 words

🤖 AI Engineering - Retrieval-Augmented Generation (RAG) Systems

⚡Direct Technical Summary

Retrieval-Augmented Generation (RAG) systems have become a crucial component in many AI applications, but they come with a significant drawback: they can hallucinate. In this artic

🤖 AI Engineering - Retrieval-Augmented Generation (RAG) Systems

Retrieval-Augmented Generation (RAG) systems have become a crucial component in many AI applications, but they come with a significant drawback: they can hallucinate. In this article, we will explore the concept of RAG systems, their limitations, and a recent breakthrough that aims to reduce hallucination.

Key Points:

  • RAG Systems: RAG systems are a type of AI model that combines retrieval and generation capabilities. They work by first retrieving relevant information from a database and then generating text based on that information.

  • Hallucination in RAG Systems: Hallucination occurs when the generated text is not based on the actual information retrieved from the database, but rather on the model's own biases and assumptions.

  • Breakthrough: IBM's RAG System: IBM has developed a RAG system that hallucinates 65x less than fine-tuned RAG systems. This breakthrough is achieved without using vector databases, embeddings, or re-rankers.

🔗 Resources:

  • Original post ↗
  • IBM
  • Retrieval-Augmented Generation (RAG) Systems
  • IBM's RAG System

🚀 Space Exploration - NASA and Mira Integration

NASA has partnered with Mira to provide a platform for users to access a wide range of space-related data, including Earth imagery, asteroids, and solar activity. This integration allows users to connect NASA to Mira via MCP and ask questions, making space exploration more accessible to everyone.

Key Points:

  • NASA and Mira Integration: The integration between NASA and Mira provides a platform for users to access a wide range of space-related data.

  • MCP: MCP is a protocol that allows users to connect NASA to Mira and ask questions.

  • Space Exploration: This integration makes space exploration more accessible to everyone, allowing users to access a wide range of space-related data.

🔗 Resources:

🌿 Chemical Perception - Scentience's Approach

Scentience has developed an approach to chemical perception that aims to bridge the gap between machine and human perception. By using aroma descriptors like "floral", "musky", and "fresh", Scentience's approach provides a more nuanced understanding of chemical reactions.

Key Points:

  • Chemical Perception: Scentience's approach to chemical perception aims to bridge the gap between machine and human perception.

  • Aroma Descriptors: Aroma descriptors like "floral", "musky", and "fresh" provide a more nuanced understanding of chemical reactions.

  • Scentience's Approach: Scentience's approach uses aroma descriptors to provide a more nuanced understanding of chemical reactions.

🔗 Resources:

🚫 AppSec - What is AppSec and What Do AppSec Engineers Do?

AppSec engineers play a crucial role in ensuring the security of software applications. In this article, we will explore what AppSec is, what AppSec engineers do, and the tools they use to ensure the security of software applications.

Key Points:

  • AppSec: AppSec is the practice of ensuring the security of software applications.

  • AppSec Engineers: AppSec engineers play a crucial role in ensuring the security of software applications.

  • SAST and DAST: SAST and DAST are two tools used by AppSec engineers to ensure the security of software applications.

🔗 Resources:

🤖 AI Engineering - System One Model

System One is an open model that matches Jev across computer use, gaming, and tool calling while running up to 9× faster. This model is designed to be used locally through the Node CLI and can be connected to the Grid to serve it across the network.

Key Points:

  • System One Model: System One is an open model that matches Jev across computer use, gaming, and tool calling.

  • 9× Faster: System One runs up to 9× faster than other models.

  • Node CLI: System One can be used locally through the Node CLI.

🔗 Resources:

📱 Intel Macs - x86_64 Support

Intel Macs are now supported alongside Apple Silicon, thanks to the recent addition of x86_64 support. This means that users can run the same code on both Intel and Apple Silicon machines.

Key Points:

  • x86_64 Support: x86_64 support has been added for Intel Macs.

  • Intel Macs: Intel Macs are now supported alongside Apple Silicon.

  • Code Compatibility: The same code can be run on both Intel and Apple Silicon machines.

🔗 Resources:

🤖 AI Engineering - Microagi's Goal

Microagi's goal is to find a way to change the universal ratio of non-running 944s. This ratio is a significant challenge in the field of AI engineering, and Microagi is working to overcome it.

Key Points:

  • Microagi's Goal: Microagi's goal is to find a way to change the universal ratio of non-running 944s.

  • Universal Ratio: The universal ratio is a significant challenge in the field of AI engineering.

  • Microagi's Approach: Microagi is working to overcome the universal ratio.

🔗 Resources:

🎁 iPhone Duo - Zoah Waitlist

Zoah is giving away an iPhone Duo to users who join their waitlist. This is a unique opportunity for users to get their hands on a new iPhone Duo before it's released to the public.

Key Points:

  • Zoah Waitlist: Zoah is giving away an iPhone Duo to users who join their waitlist.

  • iPhone Duo: The iPhone Duo is a new device that is not yet available to the public.

  • Waitlist: Users can join the waitlist to get their hands on the iPhone Duo.

🔗 Resources:

🤖 AI Engineering - Abstraction Tier

In 5 years, we won't speak about transformers as much anymore. We will have moved up the abstraction tier to organizing ecosystems. The fundamentals of transformers, agentic AI, and other principles will be important, but the higher abstraction stack will become dominant.

Key Points:

  • Abstraction Tier: In 5 years, we will have moved up the abstraction tier to organizing ecosystems.

  • Transformers: Transformers will no longer be the dominant technology in AI engineering.

  • Abstraction Stack: The higher abstraction stack will become dominant.

🔗 Resources:

🤖 AI Engineering - Model Definition

A model is not just a set of weights, but also the harness and tooling suite available to it. Our assignment of imaginary borders is largely a human one. We want there to be a delimiter between the model and the harness, but this is not always possible.

Key Points:

  • Model Definition: A model is not just a set of weights, but also the harness and tooling suite available to it.

  • Harness and Tooling Suite: The harness and tooling suite are an essential part of the model.

  • Model and Harness: The model and harness are not always separable.

🔗 Resources:

📂Source / Implementation:AI Companies and Ventures / resources-272.md
GitHub Repository↗

Related AI Companies and Ventures Breakdowns

🤖 Robotics - Compact 6-Axis Robot Arm
7m readRead ↗
Welcome Mike Hickman as Boomi's New CFO
6m readRead ↗
PyTorch Foundation OSPO & Academic Outreach Working Group Invites Academic Open Source Projects to Submit for Day 0 Workshop at PyTorchCon North America 2026
8m readRead ↗
🤖 AI Skills - Handshake's AI Skills Studio
7m readRead ↗
Drishtant Ghosh (Drix10)
Drishtant Ghosh (Drix10)•Author & Engineer

Technical founder and engineer working across AI systems, developer infrastructure, and cybersecurity.

Portfolio·GitHub·LinkedIn·X·Email
← Back to all breakdowns
GitHub: AI Companies and Ventures/resources-272.md ↗https://blogs.drix10.com/articles/ai-companies-and-ventures/ai-engineering-retrieval-augmented-generation-rag-systems-272