KATHESAMA · FIELD NOTES
Field notes from building JuanaIA
Architecture decisions, performance investigations, and the long path from remembering information to understanding it.
04 notes · English originals · Spanish local editions
Featured field note
When Your AI Has Photographic Memory But No Understanding: Designing Curiosity-Driven Knowledge Enrichment
A design for turning passive RAG storage into a governed knowledge engine that maps domains, detects gaps, and investigates them autonomously.
Series 01—04
The JuanaIA build log
Four systems notes, ordered as the project evolved from local assistant to governed knowledge engine.
- 01
I’m Building a Personal AI That Lives on My PC — Here’s What I’ve Learned So Far
A field report on building JuanaIA from scratch with local models, persistent memory, RAG, and Zero Trust service boundaries.
- 02
I Gave My Local AI a Brain: How I Designed the Orchestration Layer
Inside the Planner, Soul, and Ruflo architecture that separates decisions, execution, resilience, and auditability in JuanaIA.
- 03
From 24 Seconds to 2: How I Optimized Response Times in a Self-Hosted AI Assistant
A traced performance investigation that replaced llama.cpp with vLLM, measured the result, and added hybrid retrieval without giving up isolation.
- 04
When Your AI Has Photographic Memory But No Understanding: Designing Curiosity-Driven Knowledge Enrichment
A design for turning passive RAG storage into a governed knowledge engine that maps domains, detects gaps, and investigates them autonomously.