Kirk's Blog

Kirk's Blog

Sign in Subscribe

Optimizing PageIndex: From 10-Minute Queries to a 2.6x Speedup with Meta-Index Pre-Filtering

The table-of-contents approach: how a lightweight document index and parallel execution cut PageIndex query times by more than half – and improved recall in the process. In our first PageIndex evaluation, we found that vectorless RAG delivered impressive precision at a devastating cost: 9.6-minute average query times with Qwen3-32B. Our

stdio vs Streamable HTTP: Choosing the Right MCP Transport

stdio vs Streamable HTTP: Choosing the Right MCP Transport

A practical guide to understanding MCP's transport layer — what each option does, why SSE was deprecated, and when to use which. The Model Context Protocol has become the de facto standard for connecting AI agents to external tools and data sources. But beneath all the excitement about tools,

How Small Can You Go? Testing PageIndex with a Sub-1B Parameter Model

How Small Can You Go? Testing PageIndex with a Sub-1B Parameter Model

Pushing the lower boundary: a 0.8B parameter model cannot reliably perform PageIndex's structured tree-based retrieval, establishing a practical minimum between 1B and 4B parameters. In our previous experiment, we found that shrinking PageIndex's LLM from Qwen3-32B (20 GB) to Qwen3.5-4B (3.4 GB) improved

Does Model Size Matter for Vectorless RAG? Testing PageIndex with a 4B Parameter Model

A follow-up experiment: swapping Qwen3-32B for Qwen3.5-4B in PageIndex to measure how LLM size affects tree-based retrieval speed and quality. In our previous evaluation, we found that PageIndex – a vectorless RAG approach that uses LLM reasoning instead of embedding similarity – delivered impressive precision but at a devastating cost: 9.

Enterprise Search Types

Enterprise Search Types

When building or evaluating search for enterprise file and object stores, there are several distinct paradigms to understand: Lexical Search (Keyword / Full-Text) Matches documents based on exact or near-exact term occurrence. Uses inverted indexes (e.g., Elasticsearch, Solr, Lucene). * How it works: Tokenizes text, applies stemming/stop-word removal, scores via

Vectorless RAG with PageIndex: Can an LLM Replace Your Embedding Model?

A side-by-side comparison of tree-based LLM retrieval against three vector embedding providers, evaluated on the same corpus and queries from our earlier benchmark. The conventional RAG pipeline is well understood: chunk your documents, embed them into vectors, store them in a vector database, and at query time embed the query

Item-Level Permissions in RAG: Why Your Vector Database Needs Access Control

Item-Level Permissions in RAG: Why Your Vector Database Needs Access Control

Adding NTFS ACL enforcement to a Milvus RAG pipeline -- because retrieval quality means nothing if the wrong people see the results. In our previous article, we evaluated three embedding providers on a 72-document news corpus and found that model choice meaningfully affects retrieval quality. We spent time worrying about

Comparing Embedding Models for RAG: Ollama/Qwen3 vs Perplexity pplx-embed: A Side-by-Side Evaluation

A real-world evaluation of three embedding providers on a diverse 72-document corpus, running locally on an NVIDIA RTX 5090. When building a Retrieval-Augmented Generation (RAG) system, the embedding model is one of the most consequential decisions you will make. It determines how your documents are represented in vector space, and

Powering Search: From Keywords to Semantic Understanding

Powering Search: From Keywords to Semantic Understanding

The Evolution Beyond Traditional Search The transformation from unstructured to structured data fundamentally enables modern search capabilities that extend far beyond simple keyword matching. Traditional keyword-based search works by matching exact words in a query to those present in indexed documents, but this approach struggles with synonyms and context, misspellings

The Transformation Imperative: From Chaos to Structure

The Transformation Imperative: From Chaos to Structure

Understanding the Structural Divide The fundamental difference between structured and unstructured data lies not just in organization, but in immediate usability for business applications. Structured data is organized in a clear, predefined format, conforming to fixed schemas where data elements are categorized into rows and columns. This organization makes structured

The Unstructured Data Challenge: Volume, Complexity, and Hidden Costs

The Unstructured Data Challenge: Volume, Complexity, and Hidden Costs

Last week we introduced the hidden foundation: how unstructured data powers modern enterprise intelligence. This week, we'll be exploring the staggering numbers surrounding unstructured data. Data classified as unstructured is growing at a rate of 55-65% annually, creating an exponentially expanding universe of information that traditional tools struggle

An image depicting a modern data foundation

The hidden foundation: How structuring unstructured data powers modern enterprise intelligence

Enterprise data has reached a critical inflection point. While organizations generate vast amounts of information daily, 80-90% of this data remains unstructured, existing in formats like emails, documents, videos, and social media posts that resist traditional analysis methods. Yet beneath this seeming chaos lies extraordinary potential. McKinsey estimates that generative-AI

So You Think You Need RAG? You Probably Don’t...

So You Think You Need RAG? You Probably Don’t...

The Retrieval-Augmented Generation (RAG) boom is dazzling teams who want “ChatGPT-like” answers over their own data. Yet for many business and agentic workloads the old work-horse of lexical (keyword) search still beats RAG on speed, cost and engineering simplicity. This article explains why. RAG: Why All the Hype? RAG marries

Have you tried these reasoning techniques with Copilot?

Have you tried these reasoning techniques with Copilot?

I have spent the last few months working closely with Microsoft and some of their largest customers as they seek to implement Copilot (Office 365) into their organizations. I'll cover implementation and secure connection of enterprise data sources (i.e. non-Office 365) to Copilot in another post, but

Engage Warp, Enterprise! Generative AI and the New Frontier of Startups

Engage Warp, Enterprise! Generative AI and the New Frontier of Startups

Captain's Log, Stardate 57432.1: The winds of change are whipping through the startup galaxy, Captain. The old Starfleet adage, "bigger is better," may no longer apply to navigating the treacherous asteroid fields of venture capital. Gone are the days of phasering through pitch meetings with

Buckle Up, Breachers Be Hackin': Top 5 Cyber Security Trends for 2024

Buckle Up, Breachers Be Hackin': Top 5 Cyber Security Trends for 2024

The digital Wild West ain't tamed yet, pardners. Cyber outlaws are still slinging code and pilfering data, and 2024 promises a rootin' tootin' rollercoaster ride in the cyber security landscape. So saddle up, grab your trusty firewall, and let's explore the top 5 trends

How to: Configure Ghost email SMTP for iCloud custom domain

How to: Configure Ghost email SMTP for iCloud custom domain

Struggling to configure Ghost to use iCloud? Here’s how to do it.