From the archive
AI
Artificial intelligence, machine learning, and data science tutorials and implementations
- ↗
PostgreSQL Full-Text Search: Test the Search You Actually Need
Test PostgreSQL full-text search with a disposable database: phrases, field weights, zero-price filters, updates and the limits of a tiny fixture.
Includes verification notes - ↗
QLoRA on a Single GPU: Does a 7B Model Fit in 8GB?
Use a tested Python memory calculator to separate four-bit weights from QLoRA training memory, and measure the parts an 8GB claim leaves out.
Includes verification notes - ↗
FAISS, Pinecone and Weaviate: Compare the Search Contract First
Compare FAISS, Pinecone and Weaviate search contracts through a runnable FAISS cosine, filtering and persistence experiment.
Includes verification notes - ↗
Inside Elasticsearch: Terms, Positions and the Search You Asked For
Trace terms, positions and BM25 with a tested Python model, then separate those mechanics from Elasticsearch analysis, refresh and shard behavior.
Includes verification notes - ↗
Build RAG From Scratch in Python: Test Retrieval With LangChain
Build a tested LangChain retrieval-to-prompt pipeline with source IDs, access scopes and repeatable indexing, before adding model generation.
Includes verification notes - ↗
ngrok With Local LLMs: Test Authentication Before Sharing Ollama
Test the HTTP authentication boundary before sharing Ollama or LM Studio through ngrok, with a runnable local preflight and explicit integration limits.
Includes verification notes - ↗
Grok-1's Mixture of Experts: Active Parameters Are Not Stored Parameters
Understand Grok-1 sparse routing with an executable eight-expert NumPy experiment that separates active computation from stored weights.
Includes verification notes - ↗
Timing LLM inference in Python without inventing a CUDA speedup
Build a tested inference timer around an offline GPT-2 fixture, count generated tokens correctly and state exactly what a CUDA measurement would include.
Includes verification notes - ↗
How to Train an LLM: Verify the Training Step Before Scaling It
Run a tiny Hugging Face language model on CPU and test label alignment, padding and saved weights while training and holdout losses diverge.
Includes verification notes - ↗
Medical marketing meets a full booking queue
A tested Python queue model shows how booking capacity changes a medical marketing funnel, and where its synthetic assumptions stop.
Includes verification notes - ↗
Machine Learning in Ruby: Train, Check and Reload a Ruby-FANN Network
Train an actual Ruby-FANN XOR network, validate inputs and reload saved weights using a documented Ruby 3.3 compatibility build.
Includes verification notes - ↗
Healthcare ML: a perfect score can be a broken split
A reproducible Python example shows patient leakage turning memorization into perfect accuracy, then checks how prevalence changes positive predictions.
Includes verification notes