Pondhouse Data Blog

blog preview

BigQuery Text-to-SQL: Query Your Database in Natural Language

Build a text-to-SQL agent for BigQuery in Python: GPT-4o writes the SQL, errors are fed back for self-correction, and RAG picks relevant columns.

11 min read
Read Post
blog preview

How to finetune LLMs

Finetune Llama 3 8B on a single GPU with QLoRA, bitsandbytes and TRL: dataset formats, LoRA and training parameters, full Python script and inference.

13 min read
Read Post
blog preview

LLM Document Extraction: Pull Structured Data from Complex PDFs

Extract text and tables from complex PDFs with an open-source vision LLM: render each page as an image and prompt MiniCPM-Llama3-V 2.5. Python code included.

8 min read
Read Post
blog preview

Using AI directly from your database - with PostgreSQL and pgai

Call OpenAI from PostgreSQL with the pgai extension: create embeddings, summaries and content moderation in plain SQL, plus triggers that embed new rows.

10 min read
Read Post
blog preview

KAN: Is this the end of feedforward networks?

How Kolmogorov-Arnold Networks (KANs) replace MLP weights with learnable splines, what the paper shows on accuracy, and why MLPs aren't done yet.

9 min read
Read Post
blog preview

LLM Safety with Llama Guard 2

Llama Guard 2 is Meta's 8B model for checking LLM prompts and responses. The main attack vectors, its 11 risk categories and how to run it.

9 min read
Read Post
blog preview

How to use AI to create a PowerPoint presentation?

Let GPT-4 write python-pptx code that builds a PowerPoint deck from your company documents via RAG, using your own corporate template.

7 min read
Read Post
blog preview

Vertex AI Agent Builder: What It Is, with a Slack Search Example

What Vertex AI Agent Builder is and what Google calls it today, with a no-code example: index Slack messages and build a Gemini search app.

10 min read
Read Post
blog preview

BitNet: LLM Quantization at its Extreme

How Microsoft's BitNet trains LLMs with 1-bit and ternary weights, why it beats post-training quantization, and what it saves in memory and energy.

10 min read
Read Post
blog preview

Chat with your Confluence: A Step-by-Step Guide using Airbyte, Langchain and PGVector

Build a Confluence chatbot in Python: load pages with PyAirbyte and LangChain's AirbyteLoader, store embeddings in PGVector, answer with GPT-4 Turbo.

8 min read
Read Post