AI Augmented Investigative Journalism

Initial explorations in using current AI tools and methods to supplement existing investigative journalism techniques

Local AI Assisted Development with OpenCode, Qwen 3.6 and llama.cpp

With release of Qwen 3.6 35B A3B, AI assisted development is now useful on local setup with limited VRAM.

Processing Scanned Documents with AI

Older document processing methods still need to be used in conjunction with current AI Visual Language Models (VLMS) when dealing with badly scanned documents.

RAG with newer 30B models, Google Pinpoint and VLMs

LightRAG with Qwen3 30B Thinking model, Google Pinpoint and VLMs to speed up journalism workflows

Cleaning up Documents with VLMs

Cleaning up documents, this time with VLMs and AI generated one off scripts

Querying Federation of Malaya Budget from 1959

Current state of VLMs allows us to query accurately printed tables from the late 1950's

Local AI with AMD Radeon 9070 XT on Ubuntu Linux 25.04 with ROCm 6.4.1

Everything seems to be working so far

ROCm and Vulkan on Ubuntu VM on FreeBSD with Bhyve GPU Passthrough

With FreeBSD 15 and bhyve hypervisor you can can run and serve GPU powered local AI and compute services such as llama.cpp using ROCm and Vulkan with AMD GPUs ...