IT Expert Blog — since 2017

Practical knowledge on AI and technology for every IT Pro

Articles on local AI models, Python programming, developer tools and IT project management. No fluff — just what works in practice.

Browse Articles 🤖 AI Category

110+ articles

6 categories

2017 since

Artur Poniedziałek

IT Expert & Project Manager

AI enthusiast, local LLMs and practical tech solutions. Managing IT projects and sharing experience on this blog.

🤖 AI in Practice ⚡ IT Management 🐍 Python 🖥️ Local LLMs

🤖 Local AI & LLMs 🗄️ RAG & Vector Databases ⚙️ AI Tools & Automation 🐍 Python & Data 🔧 Software Development & Tools 📈 IT Trends & Reviews

Latest Articles

Latest

🤖 Local AI & LLMs

Kev on AMD ROCm: 171 ms decision model on a Radeon PRO W7900

We ran the open-weights Kev decision models on an AMD Radeon PRO W7900 under Windows. Accuracy matched the published model cards to three decimals; speed landed in NVIDIA L4 territory.

8th September 2026 15 min

Read article

🗄️ RAG & Vector Databases

4th September 2026 17 min

RAG framework ranking on SEC filings: seven frameworks, 28 modes, one frozen scorecard

The final ranking of the series: LightRAG, GraphRAG, RAGFlow, LlamaIndex, Haystack, R2R and kotaemon on 33 SEC filings and 40 cross-document questions,…

Read more #benchmark

🗄️ RAG & Vector Databases

1st September 2026 22 min

kotaemon benchmark on one GPU: hybrid mode is full-text search, and four builds gave four indexes

kotaemon v0.12.0 on 33 SEC filings, four generations on four containers: vector, full-text and hybrid retrieval answered 12, 15 and 14 of…

R2R benchmark on one GPU: hybrid search returned semantic search, and the search limit decided 11 or 22 of 40

Four modes of R2R 3.6.6 on 33 SEC filings, each on its own freshly built pod: as shipped it answered 11 of…

Haystack benchmark on one GPU: BM25 alone beat dense search, and 200-word chunks answered 27 of 40

Five retrieval recipes of Haystack 3.1.1 on 33 SEC filings, each on its own freshly built index: BM25 alone answered 15 of…

LlamaIndex benchmark on one GPU: smaller chunks beat the reranker, and document recall picked the wrong winner

Five retrieval configurations of LlamaIndex on 33 SEC filings, each on its own freshly built index: 19 of 40 answers correct with…

RAGFlow benchmark on one GPU: RAPTOR beat the knowledge graph four to one, and neither is a retrieval mode

A measured RAGFlow benchmark on 33 SEC filings with a local 27B model: 67.5% document recall, 13 of 40 answers right with…

GraphRAG benchmark on one GPU: 268 communities, 12 hours, one answer in five

A measured GraphRAG benchmark on a local 27B model: the community graph retrieved the evidence for 100% of 40 cross-document questions and…

LightRAG benchmark on local hardware: 33 SEC filings, one knowledge graph

A measured LightRAG benchmark on a local 27B model: the graph reached the evidence for 100% of 40 cross-document questions, yet answered…

Running LLMs on Your Own Hardware: From “Will It Even Fit” to a Model That Does Your Work

Every guide to running language models locally begins with installation. Installation is not where people get stuck. They get stuck one step…

🐍 Python & Data

5th August 2026 5 min

Python for Data Work: From a Spreadsheet Somebody Sent You to a Report Somebody Reads

Most Python data tutorials solve one link in a chain. Read a CSV. Draw a chart. Write a PDF. Each one ends…

🗄️ RAG & Vector Databases

4th August 2026 48 min

RAG golden set: extracting benchmark questions with an LLM

How we turned 33 SEC filings into 40 benchmark questions that genuinely require combining documents. The model proposed 214; ten machine checks…

Read more #benchmark

🗄️ RAG & Vector Databases

31st July 2026 14 min

RAG on SEC filings: anatomy of a benchmark corpus

Before comparing seven RAG frameworks, we measured the data they will face. 33 NVIDIA filings, 749,859 tokens, and eight repeatable patterns of…

Read more #benchmark

🤖 Local AI & LLMs

29th July 2026 9 min

NVMe vs SATA LLM model loading: 10 runs per drive, measured

A 19.92 GiB Qwen model loaded 2.25x faster from NVMe than SATA, yet time to first token was identical. Cold-start benchmark with…

Read more #benchmark

🔧 Software Development & Tools

22nd July 2026 12 min

Windows won’t boot after cloning: five real traps and their fixes

Five failures from one real SATA-to-NVMe migration: the NTFS dirty bit, duplicate EFI GUIDs, a blocked Extend Volume, a disabled controller and…

Practical knowledge on AI and technology for every IT Pro

Artur Poniedziałek

Free AI Video, Generated Locally

Latest Articles